WO2020042866A1 - 线段检测方法、装置、设备和计算机可读存储介质 - Google Patents

线段检测方法、装置、设备和计算机可读存储介质 Download PDF

Info

Publication number
WO2020042866A1
WO2020042866A1 PCT/CN2019/098927 CN2019098927W WO2020042866A1 WO 2020042866 A1 WO2020042866 A1 WO 2020042866A1 CN 2019098927 W CN2019098927 W CN 2019098927W WO 2020042866 A1 WO2020042866 A1 WO 2020042866A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
gradient
line segment
seed
pixels
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/098927
Other languages
English (en)
French (fr)
Inventor
张欢欢
刘童
唐小军
张忆非
张治国
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
BOE Technology Group Co Ltd
Original Assignee
BOE Technology Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by BOE Technology Group Co Ltd filed Critical BOE Technology Group Co Ltd
Priority to EP19850806.1A priority Critical patent/EP3846120A4/en
Priority to US16/641,830 priority patent/US11276178B2/en
Publication of WO2020042866A1 publication Critical patent/WO2020042866A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing

Definitions

  • the present application relates to the field of image processing, and in particular, to a method, an apparatus, a device, and a computer-readable storage medium for detecting a line segment.
  • a line segment is a straight line connecting two endpoints.
  • a line segment in a digital image is composed of line segment primitives connected to each other. It has simple geometric characteristics and good geometric resolution. Therefore, a line segment is a way to describe the edge characteristics of a target. .
  • the accuracy of line segment feature extraction directly affects the success or failure of subsequent image processing steps such as object recognition, stereo matching, and target tracking.
  • a line segment detection method including: extracting image features of an image to be processed, the image features including an image gradient angle and an image gradient amplitude; determining a plurality of seed points based on the extracted image features; For each seed point of the plurality of seed points, a current connected region of each seed point is determined according to a region growing rule, where the region growing rule needs to satisfy both a gradient amplitude growth rule and a gradient angle growth rule; and for each seed, Line segment fitting is performed on the line segments in the current connected area of the points.
  • a line segment detection device including: an image feature extraction unit configured to extract image features of an image to be processed, the image features including an image gradient angle and an image gradient amplitude; a seed point determination unit , Configured to determine a plurality of seed points based on the extracted image features; a region growth rule determination unit configured to determine, for each seed point of the plurality of seed points, a current connected region of each seed point according to the region growth rule, where The region growth rule needs to satisfy both the gradient amplitude growth rule and the gradient angle growth rule; and the line segment fitting unit is configured to perform line fitting on the line segments in the current connected region of each seed point.
  • a line segment detection device including: a memory configured to store computer-readable instructions; and a processor configured to process the computer-readable instructions stored in the memory, wherein all The processor performs the following functions when processing the computer-readable instructions: extracting image features of an image to be processed, the image features including an image gradient angle and an image gradient amplitude; determining a plurality of seed points based on the extracted image features; Each seed point of the plurality of seed points is determined according to a region growing rule, and the current connected region of each seed point is determined, wherein the region growing rule needs to satisfy both the gradient amplitude growth rule and the gradient angle growth rule; and for each seed point, Line segment fitting in the current connected region.
  • image feature extraction based on an image gradient fusion method can better highlight edge regions, provide good feature inputs for subsequent algorithms, and thereby achieve the purpose of reducing line segment leakage detection.
  • most of the seed points filtered by the threshold belong to the pixels in the strong edge region, which provides a good starting point for subsequent regional growth, which can reduce the generation of false line segments to a certain extent, and reduce the amount of calculation for the growth of connected regions. .
  • FIG. 1 is a flowchart of a line segment detection method according to an embodiment of the present disclosure
  • FIG. 2 is a flowchart of extracting image features of an image to be processed according to an embodiment of the present disclosure
  • FIG. 3 is a schematic diagram of a line segment detection method according to an embodiment of the present disclosure.
  • FIG. 4 is a schematic diagram of a line segment detection device according to an embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram of a line segment detection device according to an embodiment of the present disclosure.
  • the present disclosure proposes a line segment detection method in a digital image.
  • the present disclosure extracts the features of the image by fusing the gradient features of the three color channels of the color image and the gradient features of its grayscale image information. Since a line segment corresponds to a connected region with approximately equal gradient amplitudes and approximately consistent gradient directions, the present disclosure proposes a gradient-consistent connected region that uses both the gradient angle and the gradient amplitude based on the gradient characteristics of the pixels to which the edge of the line segment belongs. Growth method. After the connected region grows, the present disclosure can also perform region correction on the connected region. After region modification, the present disclosure proposes a line segment fitting method using non-maximum effective pixels of image gradients in connected regions, and determines whether the line segment fit result is valid by calculating the number of error line segments in the image.
  • FIG. 1 a line segment detection method for implementing an embodiment of the present disclosure will be described with reference to FIG. 1.
  • the method can be performed by a computer.
  • step S101 image features of an image to be processed are extracted.
  • the image may be a color image or a grayscale image.
  • Image features include image gradient angle and image gradient amplitude.
  • the image gradient is a vector that indicates that the directional derivative of the image function at that point takes a maximum along that direction, where the gradient direction (gradient angle) is the direction in which the direction derivative takes the maximum value, and the modulus of the gradient (gradient amplitude) is a function The rate of change in the direction of the gradient.
  • FIG. 2 is a flowchart of extracting image features of an image to be processed according to an embodiment of the present disclosure.
  • the gradient angle value and the gradient amplitude can be extracted based on the image gradient fusion method (S203)
  • the value is used as the image feature (S204), or the color image may be grayed (S205), and then the gradient angle value and the gradient amplitude value after the grayed image are directly extracted as the image feature (S204).
  • the gradient angle value and the gradient amplitude value of the image are directly extracted as the image features (S206).
  • the following describes the method steps of extracting the gradient angle value and the gradient amplitude value as image features based on the image gradient fusion method (S203 in FIG. 2) when the image is a color image.
  • the gradient fusion method of the color image is: firstly graying the color image and extracting the grayscale gray channel gradient amplitude; next, extracting the red channel gradient, green channel gradient, and blue channel of the color image Gradient amplitude; then take the maximum of the amplitude of the red channel gradient, green channel gradient, blue channel gradient, and gray channel gradient as the fused color image gradient amplitude; and extract the fused color image gradient amplitude
  • the corresponding gradient angle is used as the gradient angle of the fused color image.
  • the RGB color mode is a color standard in the industry. It is obtained by changing the three color channels of red (R), green (G), and blue (B) and superimposing them on each other. RGB represents the colors of the three channels of red, green and blue.
  • the red channel gradient represents the gradient of the obtained red channel image, including the gradient amplitude and the gradient angle;
  • the green channel gradient represents the gradient of the obtained green channel image, including the gradient amplitude and the gradient angle;
  • the blue channel gradient represents the obtained The gradient of the blue channel image, including gradient magnitude and gradient angle.
  • the red channel gradient, green channel gradient, and blue channel gradient of a color image reflect the local color change of the image, and the gradient of the gray image of the color image reflects the local change of image brightness.
  • a gradient fusion method can be used to extract color image features That is, the feature extraction is performed by fusing the three channels of the color image (red channel gradient, green channel gradient, and blue channel gradient) and the gradient of the grayscale gray image, so that the edge regions can be better highlighted. Provide good feature input for subsequent algorithms, and then achieve the purpose of reducing line segment leakage detection.
  • the gradient magnitude and gradient angle of the red channel of the pixel (x, y) in the color image are G r (x, y) and ⁇ r (x, y)
  • the gradient magnitude and gradient angle of the green channel are respectively G g (x, y) and ⁇ g (x, y)
  • the gradient amplitude and gradient angle of the blue channel are G b (x, y) and ⁇ b (x, y), respectively.
  • the gradient amplitude and gradient angle of the gray channel are G gray (x, y) and ⁇ gray (x, y), respectively.
  • the gradient amplitude of the color image after the gradient fusion is the maximum value of the gradient amplitude of each image, that is:
  • G fusion (x, y) max (G r (x, y), G g (x, y), G b (x, y), G gray (x, y))
  • the gradient angle after color image gradient fusion is the gradient angle corresponding to the maximum value of the gradient amplitude, that is:
  • the fused color image gradient amplitude comprehensively reflects the changes in color and brightness.
  • the fused color image gradient angle is more consistent in the color or brightness change area, thereby improving the ability to represent edge features and effectively reducing line segment leakage detection.
  • the color image can also be grayed out (S205 in FIG. 2) and directly extracted
  • the gradient angle value and gradient amplitude value after image graying are used as image features.
  • the grayscale channel ’s gradient amplitude and gradient angle are G gray (x, y) and ⁇ gray (x, y)
  • the gradient amplitude and gradient angle are directly extracted as G gray. (x, y) and ⁇ gray (x, y) as image features.
  • the gradient angle value and the gradient amplitude value of the image are directly extracted as the image features (S206 in FIG. 2).
  • step S102 a plurality of seed points are determined based on the extracted image features.
  • Determining multiple seed points based on the extracted image features may include: determining multiple seed points through non-maximum suppression and threshold filtering based on the extracted image features.
  • the image gradient reflects the sharp and continuous change of the image color or brightness.
  • the non-maximum suppression of the image gradient is the non-maximum suppression of the image gradient amplitude along the image gradient direction, so the effective pixels after the image gradient non-maximum suppression It is a candidate point for the edge of the line segment. Here, it can be used as a candidate point for the seed point growing in the connected region.
  • the non-maximum value suppression formula for image gradients is as follows. If the gradient amplitude of the current pixel is greater than the gradient amplitude of two adjacent pixels in the gradient direction, the current point gradient amplitude is retained, otherwise it is suppressed. Current point:
  • G NMS (x, y) represents the effective pixel gradient of the image gradient after non-maximum suppression
  • G (x, y) represents the image gradient amplitude of the pixel (x, y)
  • G ⁇ (x, y) 1 and G ⁇ (x, y) 2 respectively represent the magnitude of the gradient of two pixels adjacent to the current pixel in the image gradient direction. That is, if the gradient amplitude G (x, y) of the current pixel (x, y) is greater than the gradient amplitudes (G ⁇ (x, y) 1 and G ⁇ (x, y) of two adjacent pixels in the gradient direction 2 ) When both are large, the current point gradient amplitude is kept, otherwise the current point is suppressed.
  • multiple seed points are determined through threshold filtering. The process of determining multiple seed points through threshold filtering is described below.
  • the difference between the statistical average value of the gradient amplitude of the image and the statistical variance of the image gradient at a preset multiple is a threshold value.
  • a multiple of 0.2 is used as an example. All effective pixels are used as seed points for the growth of the connected area, and all pixels with gradient magnitudes greater than the sum of the statistical mean of the gradient amplitude of the image and the statistical variance of the image gradient are used as seed points for the growth of the connected area, as shown in the following formula:
  • P seed (x, y) ⁇ P (x, y)
  • threshold seed1 aveMag-0.2stdMag
  • threshold seed2 aveMag + stdMag
  • P seed (x, y) represents the set of seed points
  • P (x, y) represents the image pixels
  • G (x, y) represents the image gradient amplitude of the pixels (x, y)
  • N is the number of image pixels
  • G NMS (x, y) represents the effective pixel gradient of the image gradient after non-maximum suppression
  • aveMag represents the statistical mean value of the gradient amplitude
  • stdMag represents the statistical variance of the image gradient.
  • step S103 for each seed point of the multiple seed points, the current connected region of each seed point is determined according to the region growing rule, where the region growing rule needs to satisfy the gradient amplitude at the same time. Growth rules and gradient angle growth rules.
  • determining the current connected area of each seed point according to the region growth rule may include: selecting the seed points as the starting growth point in order from the gradient amplitude in order; For each initial growth point, other seed points that satisfy the region growth rule are added to the initial connected area of the seed point as the current connected area.
  • the region growing algorithm is a greedy algorithm, and the selection of seed points will affect the region growing result, therefore in this disclosure, in When the area grows, the seed points are selected as the initial growth points in order from the gradient amplitude in descending order, which is beneficial to reduce the occurrence of false line segments.
  • the pixels in the connected region have similar gradient directions and similar gradient amplitudes.
  • the present disclosure proposes a connected region growth rule that needs to satisfy both the gradient amplitude growth rule and the gradient angle growth rule.
  • the gradient amplitude growth rule is: sort the amplitudes of the pixels in the neighborhood of each initial growth point, and select the pixels whose amplitudes are arranged in a predetermined order as the pixels that meet the gradient amplitude growth rules. For example, the gradient magnitudes of the eight neighborhood pixels of the current growth point are sorted. If the gradient magnitude of a certain eight neighborhood pixel is ranked in the top 4, the pixel meets the gradient magnitude growth rule. It should be recognized that the above predetermined ranking is only an example, and the above predetermined ranking can be set as required.
  • the gradient angle growth rule is: pixels that are not added to the current connected area are selected as pixels to be determined, and pixels whose absolute value of the difference between the gradient angle and the gradient angle of the current connected area is less than a predetermined value are selected as pixels that satisfy the gradient angle growth rule .
  • the image gradient angle growth rule is the same as the LSD (Line Segmentation Detector,) straight line extraction algorithm, that is, if the absolute value of the difference between the gradient angle of the pixel to be determined and the gradient angle of the grown area is less than 22.5 degrees, it meets the gradient angle growth rule. . It should be recognized that the above gradient angle difference is just an example, and the above gradient angle difference can be set as needed.
  • area correction may be performed on the current connected area.
  • the area correction includes: comparing the ratio and setting of pixels in the connected area to pixels in the circumscribed rectangle. If the proportion of pixels in the connected area to the pixels of its circumscribed rectangle is less than the set proportion threshold, the area growth is restarted by reducing the image gradient angle threshold or reducing the length of the circumscribed rectangle.
  • the connected area is valid. If the proportion of the pixels in the connected area to the pixels of its circumscribed rectangle is greater than the set ratio threshold, the connected area is valid, otherwise it is invalid. If it is invalid, you need to reduce the gradient angle threshold of the image or reduce the length of the edge of the circumscribed rectangle, and then perform the region growth again to achieve the correction of the region.
  • a set threshold for example, 0.6
  • step S104 line segment fitting is performed on line segments in the current connected region of each seed point.
  • the pixels in the connected region have gradient consistency, and the non-maximum value of the image gradient suppresses the effective pixels as candidate points of the edge.
  • the line segments fitted by them are closer to the true line segment edge of the image. For this reason, this disclosure proposes a connected region Line segment fitting algorithm. First, the non-maximum value of the image is used to suppress the effective pixels to fit the line to which the line belongs, and then the image coordinate information of the connected area pixels is used to determine the two endpoints of the fitted line to determine the fitted line.
  • the image coordinates of the non-maximum value suppression effective pixel of the image gradient in the connected region are used as the input of the line fitting to the line segment, and then the least line method is used to fit the line to which the line segment belongs.
  • the two end points of the line segment are determined according to the maximum and minimum values of the image coordinates of the pixels in the connected area, and a line segment detection result is obtained. Next, it is verified whether the line segment detection result is valid.
  • the following method can be used to verify whether the line segment detection result is valid. For a line segment fitted by n seed points, calculate the distance from each seed point to the line segment, and count the number of seed point pixels whose distance is less than a predetermined value (for example, 1.5 pixels), and then calculate the image based on the line segment The number of faulty line segments (NFA).
  • a predetermined value for example, 1.5 pixels
  • NFA NFA
  • NFA (L, x) ⁇ the line segment detection result is valid; otherwise, the line segment detection result is invalid.
  • ⁇ can be set to 1, that is, an error detection result is allowed in an image. It can be understood that ⁇ can also be set to other values, and the value set for ⁇ is only used as an example here.
  • FIG. 3 is a schematic diagram of a line segment detection method according to an embodiment of the present disclosure.
  • the present disclosure proposes a line segment detection method in a digital image.
  • the image features of the image to be processed are first extracted, where the image features include the image gradient angle and the image gradient amplitude.
  • the input image is a color image 31, it can be extracted based on the method of image gradient fusion 32, or the color image can be grayscaled, converted into a grayscale image 33, and then the gradient angle value and Gradient amplitude values (for example, image gradient 34) are used as image features.
  • the input image is a grayscale image 33
  • features of the image for example, image gradient 34
  • the connected region extraction 35 is performed.
  • connected area extraction includes three aspects: first, based on the extracted image features, non-maximum suppression and threshold filtering are used to select seed points 36; next, region growth 37 is performed, that is, according to the gradient amplitude from large to small In order, the seed points are selected as the initial growth points. For each initial growth point, other seed points that satisfy the region growth rule are added to the initial connected region of the seed point as the current connected region. Among them, the region growth rule requires At the same time, the gradient amplitude growth rule and the gradient angle growth rule are satisfied; then, after the current connected area of each seed point is determined according to the area growth rule, the current connected area may be subjected to area correction 38.
  • line segment fitting 39 can be performed on the line segments in the connected region.
  • Line segment fitting includes using non-maximum image to suppress effective pixels to fit the line to which the line belongs, and then combine the image coordinate information of the pixels in the connected region. Determine the two endpoints of the fitted line segment to determine the fitted line segment.
  • the error control 40 is used to determine whether the line segment fitting result is valid, that is, the number of error line segments in the image can be calculated to determine whether the line segment fitting result is valid. Finally, the line segment detection result 41 is output.
  • edge regions can be better highlighted, good feature input can be provided for subsequent algorithms, and the purpose of reducing line segment leakage detection can be achieved.
  • most of the seed points filtered by the threshold belong to the pixels in the strong edge region, which provides a good starting point for subsequent regional growth, which can reduce the generation of false line segments to a certain extent, and reduce the amount of calculation for the growth of connected regions. .
  • selecting the seed points as the initial growth points in order of the magnitude of the gradient during the region growth is beneficial to reducing the occurrence of false line segments.
  • FIG. 4 is a schematic diagram of a line segment detection device according to an embodiment of the present disclosure. Since the function of the line segment detection device of this embodiment is the same as the details of the method described above with reference to FIG. 1, a detailed description of the same content is omitted here for simplicity.
  • the line segment detection device 1000 includes an image feature extraction unit 1001, a seed point determination unit 1002, an area growth rule determination unit 1003, an area correction unit 1004, a line segment fitting unit 1005, and a determination unit 1006. It should be noted that although the line segment detection device 1000 is shown in FIG. 4 as including only 6 units, this is only schematic. The line segment detection device 1000 may also include one or more other units. These units are related to the inventive concept. None, so it is omitted here.
  • the image feature extraction unit 1001 may extract image features of an image to be processed.
  • the image may be a color image or a grayscale image.
  • Image features include image gradient angle and image gradient amplitude.
  • the gradient angle value and the gradient amplitude value can be extracted as image features based on the image gradient fusion method, or the color image can be grayed out, and then the gradient angle value after the grayed image is directly extracted And gradient magnitude values as image features.
  • the gradient angle value and the gradient amplitude value of the image are directly extracted as the image features.
  • the following describes the method steps of extracting a gradient angle value and a gradient amplitude value as image features based on an image gradient fusion method when the image is a color image.
  • the gradient fusion method of the color image is: firstly graying the color image and extracting the grayscale gray channel gradient amplitude; next, extracting the red channel gradient, green channel gradient, and blue channel of the color image Gradient amplitude; then take the maximum of the amplitude of the red channel gradient, green channel gradient, blue channel gradient, and gray channel gradient as the fused color image gradient amplitude; and extract the fused color image gradient amplitude
  • the corresponding gradient angle is used as the gradient angle of the fused color image.
  • the red channel gradient, green channel gradient, and blue channel gradient of a color image reflect the local color change of the image, and the gradient of the gray image of the color image reflects the local change of image brightness.
  • a gradient fusion method can be used to extract color image features That is, the feature extraction is performed by fusing the three channels of the color image (red channel gradient, green channel gradient, and blue channel gradient) and the gradient of the grayscale gray image, so that the edge regions can be better highlighted.
  • the edge regions can be better highlighted.
  • the gradient magnitude and gradient angle of the red channel of the pixel (x, y) in the color image are G r (x, y) and ⁇ r (x, y)
  • the gradient magnitude and gradient angle of the green channel are respectively G g (x, y) and ⁇ g (x, y)
  • the gradient amplitude and gradient angle of the blue channel are G b (x, y) and ⁇ b (x, y), respectively.
  • the gradient amplitude and gradient angle of the gray channel are G gray (x, y) and ⁇ gray (x, y), respectively.
  • the gradient amplitude of the color image after the gradient fusion is the maximum value of the gradient amplitude of each image, that is:
  • G fusion (x, y) max (G r (x, y), G g (x, y), G b (x, y), G gray (x, y))
  • the gradient angle after color image gradient fusion is the gradient angle corresponding to the maximum value of the gradient amplitude, that is:
  • the fused color image gradient amplitude comprehensively reflects the changes in color and brightness.
  • the fused color image gradient angle is more consistent in the color or brightness change area, thereby improving the ability to represent edge features and effectively reducing line segment leakage detection.
  • the color image can also be grayed out, and the grayed-out gradient of the image can be directly extracted Angle values and gradient amplitude values are used as image features.
  • the grayscale channel ’s gradient amplitude and gradient angle are G gray (x, y) and ⁇ gray (x, y)
  • the gradient amplitude and gradient angle are directly extracted as G gray. (x, y) and ⁇ gray (x, y) as image features.
  • the gradient angle value and the gradient amplitude value of the image are directly extracted as the image features.
  • the seed point determination unit 1002 may determine a plurality of seed points based on the extracted image features.
  • the seed point determination unit 1002 determines a plurality of seed points through non-maximum value suppression and threshold filtering based on the extracted image features.
  • the image gradient reflects the sharp and continuous change of the image color or brightness.
  • the non-maximum suppression of the image gradient is the non-maximum suppression of the image gradient amplitude along the image gradient direction, so the effective pixels after the image gradient non-maximum suppression It is a candidate point for the edge of the line segment. Here, it can be used as a candidate point for the seed point growing in the connected region.
  • the non-maximum value suppression formula for image gradients is as follows. If the gradient amplitude of the current pixel is greater than the gradient amplitude of two adjacent pixels in the gradient direction, the current point gradient amplitude is retained, otherwise it is suppressed. Current point:
  • G NMS (x, y) represents the effective pixel gradient of the image gradient after non-maximum suppression
  • G (x, y) represents the image gradient amplitude of the pixel (x, y)
  • G ⁇ (x, y) 1 and G ⁇ (x, y) 2 respectively represent the magnitude of the gradient of two pixels adjacent to the current pixel in the image gradient direction. That is, if the gradient amplitude G (x, y) of the current pixel (x, y) is greater than the gradient amplitudes (G ⁇ (x, y) 1 and G ⁇ (x, y) of two adjacent pixels in the gradient direction 2 ) When both are large, the current point gradient amplitude is kept, otherwise the current point is suppressed.
  • multiple seed points are determined through threshold filtering. The process of determining multiple seed points through threshold filtering is described below.
  • the difference between the statistical average value of the gradient amplitude of the image and the statistical variance of the image gradient of 0.2 times is a threshold value, and all effective pixel points whose effective pixel gradient after the image gradient is suppressed by a non-maximum value greater than the threshold value are used as connected regions
  • the growing seed point, and all the effective pixel points whose gradient amplitude is greater than the sum of the image gradient amplitude statistical mean and the image gradient statistical variance are used as the seed points for growing the connected region, as shown in the following formula:
  • P seed (x, y) ⁇ P (x, y)
  • threshold seed1 aveMag-0.2stdMag
  • threshold seed2 aveMag + stdMag
  • P seed (x, y) represents the set of seed points
  • P (x, y) represents the image pixels
  • G (x, y) represents the image gradient amplitude of the pixels (x, y)
  • N is the number of image pixels
  • G NMS (x, y) represents the effective pixel gradient of the image gradient after non-maximum suppression
  • aveMag represents the statistical mean value of the gradient amplitude
  • stdMag represents the statistical variance of the image gradient.
  • the region growing rule determining unit 1003 determines a current connected region of each seed point according to the region growing rule, where the region growing rule needs to satisfy both the gradient amplitude growth rule and Gradient angle growth rules.
  • the region growing rule determination unit 1003 sequentially selects the seed points as the starting growth points in order of the magnitude of the gradient; for each starting growth point, it will satisfy The other seed points of the region growing rule are added to the initial connected region of the seed point as the current connected region.
  • the pixels in the connected region have similar gradient directions and similar gradient amplitudes.
  • the present disclosure proposes a connected region growth rule that needs to satisfy both the gradient amplitude growth rule and the gradient angle growth rule.
  • the gradient amplitude growth rule is: sort the amplitudes of the pixels in the neighborhood of each initial growth point, and select the pixels whose amplitudes are arranged in a predetermined order as the pixels that meet the gradient amplitude growth rules. For example, the gradient magnitudes of the eight neighborhood pixels of the current growth point are sorted. If the gradient magnitude of an eight neighborhood pixel is ranked in the top 4, the pixel meets the gradient magnitude growth rule. It should be recognized that the above predetermined ranking is only an example, and the above predetermined ranking can be set as required.
  • the gradient angle growth rule is: pixels that are not added to the current connected area are selected as pixels to be determined, and pixels whose absolute value of the gradient angle difference between the gradient angle and the current connected area is less than a predetermined value are selected as pixels that satisfy the gradient angle growth rule .
  • the image gradient angle growth rule is the same as the LSD straight line extraction algorithm, that is, if the absolute value of the difference between the gradient angle of the pixel to be determined and the gradient angle of the grown area is less than 22.5 degrees, it meets the gradient angle growth rule. It should be recognized that the above gradient angle difference is just an example, and the above gradient angle difference can be set as needed.
  • the area correction unit 1004 may perform area correction on the current connected area, and the area correction includes comparing pixels in the connected area to pixels of its circumscribed rectangle. Proportion and set proportion threshold; if the proportion of pixels in the connected area to the pixels of the circumscribed rectangle is smaller than the set proportion threshold, the region growth is restarted by reducing the image gradient angle threshold or reducing the length of the circumscribed rectangle.
  • the line segment fitting unit 1005 performs line segment fitting on line segments in the current connected region of each seed point. That is, the non-maximum value of the image can be used to suppress the effective pixels to fit the line to which the line belongs, and then the two end points of the fitted line are determined by combining the image coordinate information of the pixels in the connected area to determine the fitted line.
  • the judgment unit 1006 can determine whether the line segment fitting result is valid by calculating the number of error line segments in the image, that is, after performing line fitting on the line segments in the current connected region of each seed point, The method includes the following steps: calculating the number of error line segments in the image, and selecting a line segment fitting result whose number of error line segments is less than a preset threshold as a valid result.
  • the following method can be used to verify whether the line segment detection result is valid. For a line segment fitted by n seed points, calculate the distance from each seed point to the line segment, and count the number of pixels of the seed point with a distance less than 1.5, and then calculate the number of error line segments in the image based on the line segment (NFA ). As described above, the NFA judgment method is not repeated here.
  • FIG. 5 is a schematic diagram of a line segment detection device according to an embodiment of the present disclosure. Since the function of the line segment detection device of this embodiment is the same as the details of the method described above with reference to FIG. 1, a detailed description of the same content is omitted here for simplicity.
  • the line segment detection device 1100 includes a memory 1101 configured to store computer-readable instructions; and a processor 1102 configured to process the computer-readable instructions stored in the memory, wherein the processor processes
  • the computer-readable instructions perform the following functions: extracting image features of an image to be processed; determining a plurality of seed points based on the extracted image features; and for each of the plurality of seed points, determining each of the plurality of seed points according to a region growing rule The current connected region of the seed points, where the region growth rule needs to satisfy both the gradient amplitude growth rule and the gradient angle growth rule; and line segment fitting is performed on the line segments in the current connected region of each seed point.
  • the present disclosure also provides a computer-readable storage medium having stored thereon a computer program that, when executed by a processor, implements the following method: extracting image features of an image to be processed, the image features including an image gradient angle and an image gradient Amplitude; determine multiple seed points based on the extracted image features; for each seed point in the multiple seed points, determine the current connected region of each seed point according to the region growing rule, where the region growing rule needs to satisfy the gradient amplitude at the same time Growth rules and gradient angle growth rules; and line segment fitting of line segments in the current connected region of each seed point.
  • aspects of this application can be illustrated and described through several patentable categories or situations, including any new and useful process, machine, product or substance combination, or any of them New and useful improvements. Accordingly, various aspects of the present application can be executed entirely by hardware, can be executed entirely by software (including firmware, resident software, microcode, etc.), and can also be executed by a combination of hardware and software.
  • the above hardware or software can be called “data block”, “module”, “engine”, “unit”, “component” or “system”.
  • aspects of the present application may manifest as a computer product located in one or more computer-readable media, the product including a computer-readable program code.
  • Each unit in this application may include a microprocessor, a digital signal processor (DSP, Digital Signal Processor), an application specific integrated circuit (ASIC, Application Specific Integrated Circuit), a programmable logic device (PLD, Programmable Logic Device), and a central processing unit.
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • PLD Programmable Logic Device
  • CPU Central Processing Unit
  • GPU Graphics Processing Unit
  • FPGA Field Programmable Gate Array

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

本发明提供了一种线段检测方法、装置、设备和计算机可读存储介质。所述方法包括：提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；基于提取的图像特征确定多个种子点；对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及对各个种子点的当前连通区域内的线段进行线段拟合。

Description

线段检测方法、装置、设备和计算机可读存储介质
相关申请的交叉引用
本专利申请要求于2018年8月30日递交的中国专利申请第201811005060.7号的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本申请涉及图像处理领域，并具体涉及线段检测方法、装置、设备和计算机可读存储介质。
背景技术
根据视觉计算理论，人眼对物体的辨识首先要得到它的轮廓，计算机视觉系统要模仿人的视觉识别图像中目标首先要得到表征目标边缘轮廓的要素图。线段是连接两个端点之间笔直的线，数字图像中的线段由彼此相连的线段基元组成，其具有简单的几何特征和良好的几何解析性，因此线段是描述目标边缘特征的一种方式。对于许多计算机视觉系统来说，线段特征提取的准确程度直接影响着物体识别、立体匹配、目标跟踪等后续图像处理环节的成败。
发明内容
根据本公开的一个方面，提供了一种线段检测方法，包括：提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；基于提取的图像特征确定多个种子点；对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及对各个种子点的当前连通区域内的线段进行线段拟合。
根据本公开的一个方面，提供了一种线段检测装置，包括：图像特征提取单元，配置为提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；种子点确定单元，配置为基于提取的图像特征确定多个种子点；区域生长规则确定单元，配置为对于所述多个种子点中的每个种子点，根 据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及线段拟合单元，配置为对各个种子点的当前连通区域内的线段进行线段拟合。
根据本公开的一个方面，提供了一种线段检测设备，包括：存储器，配置为存储计算机可读指令；以及处理器，配置为处理存储在所述存储器中的所述计算机可读指令，其中所述处理器处理所述计算机可读指令时执行以下功能：提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；基于提取的图像特征确定多个种子点；对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及对各个种子点的当前连通区域内的线段进行线段拟合。
在本公开的上述方面中，通过基于图像梯度融合方法进行图像特征提取，可以更好的突显边缘区域，为后续算法提供良好特征输入，进而达到减少线段漏检测的目的。此外，经阈值筛选后的种子点绝大部分属于强边缘区域像素，这为后续的区域生长提供了良好的起始点，可在一定程度上减少虚假线段的产生，同时减少连通区域生长的计算量。
附图说明
通过结合附图对本公开实施例进行更详细的描述，本公开的上述以及其它目的、特征和优势将变得更加明显。附图用来提供对本公开实施例的进一步理解，并且构成说明书的一部分，与本公开实施例一起用于解释本公开，并不构成对本公开的限制。在附图中，相同的参考标号通常代表相同部件或步骤。
图1是根据本公开实施例的线段检测方法的流程图；
图2是根据本公开实施例的提取待处理图像的图像特征的流程图；
图3是根据本公开的实施例线段检测方法的示意图；
图4是根据本公开实施例的线段检测装置的示意图；
图5是根据本公开实施例的线段检测设备的示意图。
具体实施方式
下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述。显然，所描述的实施例仅是本公开一部分的实施例，而不 是全部的实施例。基于本公开中的实施例，本领域普通技术人员在无需创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
本公开提出了一种数字图像中的线段检测方法。为更加综合的突显图像局部亮度或颜色的变化，本公开通过融合彩色图像的三个颜色通道的梯度特征和其灰度图像信息的梯度特征来提取图像的特征。由于一个线段对应一个梯度幅值近似相等、梯度方向近似一致的连通区域，因此，本公开基于线段边缘所属像素的梯度特点，提出了同时利用梯度角度和梯度幅值的基于梯度一致性的连通区域生长方法。在连通区域生长之后，本公开还可以对连通区域进行区域修正。在区域修正之后，本公开提出了利用连通区域内图像梯度非极大值有效像素的线段拟合方法，并通过计算图像中错误线段的数目来判断线段拟合结果是否有效。
首先，参照图1来描述用于实现本公开实施例的线段检测方法。该方法可以由计算机执行。
如图1所示，在步骤S101中，提取待处理图像的图像特征。
该图像可以是彩色图像，或者也可以是灰度图像。图像特征包括图像梯度角度和图像梯度幅值。图像梯度是一个向量，表示图像函数在该点处的方向导数沿着该方向取得最大值，其中梯度方向(梯度角度)为方向导数取得最大值的方向，梯度的模(梯度幅值)为函数在梯度方向上的变化率。
图2是根据本公开实施例的提取待处理图像的图像特征的流程图。如图2所示，首先输入图像(S201)，然后判断该输入图像是否为彩色图像(S201)，当该图像为彩色图像时，可以基于图像梯度融合方法(S203)提取梯度角度值和梯度幅度值作为图像特征(S204)，或者可以将彩色图像进行灰度化(S205)，然后直接提取图像灰度化后的梯度角度值和梯度幅度值作为图像特征(S204)。可替换地，当该图像不是彩色图像(例如，该图像是灰度图像)时，则直接提取图像的梯度角度值和梯度幅度值作为图像特征(S206)。
下面说明当图像为彩色图像时，基于图像梯度融合方法(图2中的S203)提取梯度角度值和梯度幅度值作为图像特征的方法步骤。
这里，彩色图像的梯度融合方法为：首先将彩色图像进行灰度化，提取灰度化后的灰度通道梯度幅值；接下来，提取彩色图像的红色通道梯度、绿色通道梯度和蓝色通道梯度幅值；然后取红色通道梯度、绿色通道梯度、蓝色通道梯度以及灰度通道梯度的幅值的最大值作为融合后的彩色图像梯度幅值；以 及提取该融合后的彩色图像梯度幅值对应的梯度角度作为融合后的彩色图像梯度角度。
这里，RGB色彩模式是工业界的一种颜色标准，是通过对红(R)、绿(G)、蓝(B)三个颜色通道的变化以及它们相互之间的叠加来得到各式各样的颜色，RGB代表红、绿、蓝三个通道的颜色。红色通道梯度表示求取的红色通道图像的梯度，包括梯度幅值和梯度角度；绿色通道梯度表示求取的绿色通道图像的梯度，包括梯度幅值和梯度角度；蓝色通道梯度表示求取的蓝色通道图像的梯度，包括梯度幅值和梯度角度。
彩色图像的红色通道梯度、绿色通道梯度和蓝色通道梯度反映了图像局部颜色变化，彩色图像的灰度图像的梯度反映了图像亮度的局部变化，例如，可以采用梯度融合方法来提取彩色图像特征，即通过融合彩色图像的3个通道(红色通道梯度、绿色通道梯度和蓝色通道梯度)和其灰度化后的灰度图像的梯度来进行特征提取，从而可以更好的突显边缘区域，为后续算法提供良好特征输入，进而达到减少线段漏检测的目的。
例如，假设彩色图像中像素(x,y)的红色通道的梯度幅值和梯度角度分别为G r(x,y)和θ r(x,y)，绿色通道的梯度幅值和梯度角度分别为G g(x,y)和θ g(x,y)，蓝色通道的梯度幅值和梯度角度分别为G b(x,y)和θ b(x,y)，灰度化后的灰度通道的梯度幅值和梯度角度分别为G gray(x,y)和θ gray(x,y)，则彩色图像梯度融合后的梯度幅值为各图像梯度幅值的最大值，即：
G fusion(x,y)＝max{G r(x,y),G g(x,y),G b(x,y),G gray(x,y)}
彩色图像梯度融合后的梯度角度为梯度幅值最大值对应的梯度角度，即：
Figure PCTCN2019098927-appb-000001
融合后的彩色图像梯度幅值综合反映了颜色和亮度的变化，融合后的彩色图像梯度角度在颜色或亮度变化区域更加一致，从而提高了边缘特征的表征能力，有效地减少线段漏检测。
在说明了图像为彩色图像时，基于图像梯度融合方法提取梯度角度值和梯度幅度值作为图像特征之后，可替换地，还可以将彩色图像进行灰度化(图2中的S205)，直接提取图像灰度化后的梯度角度值和梯度幅度值作为图像特征。
例如，假设图像灰度化后的灰度通道的梯度幅值和梯度角度分别为 G gray(x,y)和θ gray(x,y)，则直接提取梯度幅值和梯度角度分别为G gray(x,y)和θ gray(x,y)作为图像特征。
或者，当该图像不是彩色图像(例如，该图像是灰度图像)时，则直接提取图像的梯度角度值和梯度幅度值作为图像特征(图2中的S206)。接下来，返回图1，在步骤S102中，基于提取的图像特征确定多个种子点。
基于提取的图像特征确定多个种子点可以包括：基于提取的图像特征，通过非极大值抑制和阈值筛选确定多个种子点。
图像梯度反映了图像颜色或亮度急剧连续变化，图像梯度的非极大值抑制是沿着图像梯度方向对图像梯度幅值进行非极大值抑制，因此图像梯度非极大值抑制后的有效像素是线段边缘的候选点，这里，可将其作为连通区域生长的种子点的候选点。
例如，图像梯度的非极大值抑制公式如下所述，即如果当前像素的梯度幅值比其在梯度方向上相邻两像素的梯度幅值都大，则保留当前点梯度幅值，否则抑制当前点：
Figure PCTCN2019098927-appb-000002
其中，G NMS(x,y)表示图像梯度经非极大值抑制后的有效像素梯度，G(x,y)表示像素(x,y)的图像梯度幅值，G θ(x,y) 1和G θ(x,y) 2分别表示在图像梯度方向上与当前像素相邻的两像素的梯度幅值。即，如果当前像素(x,y)的梯度幅值G(x,y)比其在梯度方向上相邻两像素的梯度幅值(G θ(x,y) 1和G θ(x,y) 2)都大时，保留当前点梯度幅值，否则抑制当前点。
在通过非极大值抑制确定多个有效像素梯度后，接下来，通过阈值筛选来确定多个种子点，下面介绍通过阈值筛选确定多个种子点的过程。
本公开以图像的梯度幅值统计均值与预设倍数图像梯度统计方差之间的差值为阈值，这里以0.2倍数示例，将图像梯度经非极大值抑制后的有效像素梯度大于该阈值的所有有效像素点作为连通区域生长的种子点，同时将所有梯度幅值大于该图像梯度幅值统计均值和图像梯度统计方差之和的所有像素点作为连通区域生长的种子点，如下公式所示：
P seed(x,y)＝{{P(x,y)|G NMS(x,y)>threshold seed1}U{P(x,y)|G(x,y)>threshold seed2}}
其中，
threshold seed1＝aveMag-0.2stdMag
threshold seed2＝aveMag+stdMag
aveMag＝(1/N)∑G(x,y)
Figure PCTCN2019098927-appb-000003
其中P seed(x,y)表示种子点集合，P(x,y)表示图像像素，G(x,y)表示像素(x,y)的图像梯度幅值，N为图像像素的数目，G NMS(x,y)表示图像梯度经非极大值抑制后的有效像素梯度，aveMag表示梯度幅值统计均值，stdMag表示图像梯度统计方差。
如上所述，图像梯度经非极大值抑制后的有效像素有很多，有些属于强边缘区域，但是大部分为弱边缘或噪声区域，强边缘区域像素的梯度幅值较大，弱边缘区域像素的梯度幅值较小。经阈值筛选后的种子点绝大部分属于强边缘区域像素，这为后续的区域生长提供了良好的起始点，可在一定程度上减少虚假线段的产生，同时减少连通区域生长的计算量。
接下来，继续回到图1，在步骤S103中，对于多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则。
其中，对于多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域可以包括：按照梯度幅值由大到小的顺序依次选取种子点作为起始生长点；对于每个起始生长点，将满足区域生长规则的其他种子点加入到该种子点的初始连通区域中作为当前连通区域。
梯度幅值越大表示图像的亮度或颜色变化越明显，越有可能是图像边缘区域，而由于区域生长算法为贪心算法，并且种子点的选取会影响区域生长结果，因此在本公开中，在区域生长时按照梯度幅值由大到小的顺序依次选取种子点作为初始生长点，这有利于减少虚假线段的产生。
连通区域中的像素具有相近的梯度方向以及相近的梯度幅值，本公开提出了需要同时满足梯度幅值生长规则和梯度角度生长规则的连通区域生长规则。
梯度幅值生长规则为：对每个起始生长点的邻域像素的幅值进行排序，选择幅值排在预定位次的像素作为满足梯度幅值生长规则的像素。例如，对当前 生长点的八邻域像素的梯度幅值进行排序，如果某个八邻域像素的梯度幅值排在前4名，则该像素满足梯度幅值生长规则。应该认识到，以上预定位次只是一个示例，上述预定位次可以根据需要设定。
梯度角度生长规则为：将未加入到当前连通区域的像素作为待判定像素，选择梯度角度与当前连通区域的梯度角度差值的绝对值小于预定值的待判定像素作为满足梯度角度生长规则的像素。例如，图像梯度角度生长规则与LSD(Line Segment Detector，)直线提取算法相同，即若待判定像素梯度角度与已生长区域的梯度角度差值的绝对值小于22.5度，则其满足梯度角度生长规则。应该认识到，以上梯度角度差值只是一个示例，上述梯度角度差值可以根据需要设定。
进一步，在根据区域生长规则确定各个种子点的当前连通区域后，可以对所述当前连通区域进行区域修正，所述区域修正包括：比较连通区域内像素占其外接矩形的像素的比例和设定的占比阈值；如果连通区域内像素占其外接矩形的像素的比例小于设定的占比阈值，则通过减少图像梯度角度阈值或减少外接矩形边长重新进行区域生长。
例如，可以通过比较连通区域内像素占其外接矩形的像素的比例和设定的占比阈值(例如0.6)，可判断该连通区域是否有效。如果连通区域内像素占其外接矩形的像素的比例大于设定的占比阈值，则该连通区域有效，否则无效。如果无效，则需要通过减少图像梯度角度阈值或减少外接矩形边长重新进行区域生长实现对区域的修正。
接下来，返回图1，在步骤S104中，对各个种子点的当前连通区域内的线段进行线段拟合。
连通区域内像素具有梯度一致性，图像梯度的非极大值抑制有效像素为边缘的候选点，由它们拟合出的线段更接近图像的真实线段边缘，为此本公开提出了一种连通区域线段拟合算法。首先利用图像非极大值抑制有效像素进行线段所属直线的拟合，再结合连通区域像素的图像坐标信息确定拟合线段的两个端点来确定拟合的线段。
具体地，将连通区域内图像梯度非极大值抑制有效像素的图像坐标作为线段所属直线拟合的输入，然后采用例如最小二乘法进行线段所属直线的拟合。接下来，根据连通区域像素的图像坐标的最大值和最小值确定线段的两个端点，得到一个线段检测结果，接下来验证线段检测结果是否有效。
在确定了拟合的线段之后，可以通过计算图像中错误线段的数目来判断线段拟合结果是否有效，即在对各个种子点的当前连通区域内的线段进行线段拟合之后，还包括如下步骤：计算图像中错误线段的数目，选择错误线段的数目小于预设阈值的线段拟合结果作为有效结果。
例如，可以通过如下方法来验证线段检测结果是否有效。对于一个由n个种子点拟合的线段，计算每个种子点到该线段的距离，并统计距离小于预定值(例如，1.5像素)的种子点像素的个数，然后基于该线段计算该图像中错误线段的数目(NFA)。
例如，假设L为由n个种子点拟合的线段，其中至少有x个种子点到拟合线段的距离小于1.5，那么定义NFA为
Figure PCTCN2019098927-appb-000004
其中，(M×N) 2表示在(M×N)大小的图像中潜在的线段数量。由于线段具有两个端点，并且每个端点可以位于图像的任何位置，因此总计线段数量为(M×N) 2。假定连通区域的外矩形的宽度为width，那么概率p表示种子点不是线段拟合误差点的概率，p＝3/width。
如果NFA(L，x)≤ε，线段检测结果有效，否则线段检测结果无效，其中可以将ε设置为1，也就是一幅图像中允许存在一个错误检测结果。可以理解的是，也可以将ε设置为其他数值，在此对ε设置的值仅仅作为一个示例。
下面参照图3来描述根据本公开的实施例线段检测方法。如图3所示为根据本公开的实施例线段检测方法的示意图。
本公开提出了一种数字图像中的线段检测方法。对于输入的图像，首先提取待处理图像的图像特征，其中图像特征包括图像梯度角度和图像梯度幅值。当输入图像为彩色图像31时，可以基于图像梯度融合32的方法进行提取，或者可以将彩色图像进行灰度化，转换为灰度图像33，然后直接提取图像灰度化后的梯度角度值和梯度幅度值(例如，图像梯度34)作为图像特征，当输入图像为灰度图像33时，则直接提取图像的特征(例如，图像梯度34)。在提取了图像的特征之后，进行连通区域的提取35。如图所示，连通区域提取包括三个方面：首先基于提取的图像特征根据非极大值抑制和阈值筛选选取种子点36；接下来，进行区域生长37，即按照梯度幅值由大到小的顺序依次选取种子点作为起始生长点，对于每个起始生长点，将满足区域生长规则的 其他种子点加入到该种子点的初始连通区域中作为当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；然后，在根据区域生长规则确定各个种子点的当前连通区域之后，可以对当前连通区域进行区域修正38。在连通区域提取35之后，可以对连通区域内的线段进行线段拟合39，线段拟合包括利用图像非极大值抑制有效像素进行线段所属直线的拟合，再结合连通区域像素的图像坐标信息确定拟合线段的两个端点来确定拟合的线段。在确定了拟合的线段之后，通过误差控制40来判断线段拟合结果是否有效，即可以通过计算图像中错误线段的数目来判断线段拟合结果是否有效。最后，输出线段检测结果41。
可以看出，在本公开的上述方面中，通过基于图像梯度融合方法进行图像特征提取，可以更好的突显边缘区域，为后续算法提供良好特征输入，进而达到减少线段漏检测的目的。此外，经阈值筛选后的种子点绝大部分属于强边缘区域像素，这为后续的区域生长提供了良好的起始点，可在一定程度上减少虚假线段的产生，同时减少连通区域生长的计算量。进一步，在区域生长时按照梯度幅值由大到小的顺序依次选取种子点作为初始生长点，有利于减少虚假线段的产生。
下面，参照图4描述根据本公开实施例的线段检测装置。图4是根据本公开实施例的线段检测装置的示意图。由于本实施例的线段检测装置的功能与在上文中参照图1描述的方法的细节相同，因此在这里为了简单起见，省略对相同内容的详细描述。
如图4所示，线段检测装置1000包括图像特征提取单元1001、种子点确定单元1002、区域生长规则确定单元1003、区域修正单元1004、线段拟合单元1005和判断单元1006。需要注意的是，尽管在图4中线段检测装置1000被示出为只包括6个单元，但这只是示意性的，线段检测装置1000也可以包括一个或多个其他单元，这些单元与发明构思无关，因此在这里被省略。
图像特征提取单元1001可以提取待处理图像的图像特征。
该图像可以是彩色图像，或者也可以是灰度图像。图像特征包括图像梯度角度和图像梯度幅值。
这里，当该图像为彩色图像时，可以基于图像梯度融合方法提取梯度角度值和梯度幅度值作为图像特征，或者可以将彩色图像进行灰度化，然后直接提取图像灰度化后的梯度角度值和梯度幅度值作为图像特征。可替换地，当该图 像为灰度图像时，则直接提取图像的梯度角度值和梯度幅度值作为图像特征。
下面说明当图像为彩色图像时，基于图像梯度融合方法提取梯度角度值和梯度幅度值作为图像特征的方法步骤。
这里，彩色图像的梯度融合方法为：首先将彩色图像进行灰度化，提取灰度化后的灰度通道梯度幅值；接下来，提取彩色图像的红色通道梯度、绿色通道梯度和蓝色通道梯度幅值；然后取红色通道梯度、绿色通道梯度、蓝色通道梯度以及灰度通道梯度的幅值的最大值作为融合后的彩色图像梯度幅值；以及提取该融合后的彩色图像梯度幅值对应的梯度角度作为融合后的彩色图像梯度角度。彩色图像的红色通道梯度、绿色通道梯度和蓝色通道梯度反映了图像局部颜色变化，彩色图像的灰度图像的梯度反映了图像亮度的局部变化，例如，可以采用梯度融合方法来提取彩色图像特征，即通过融合彩色图像的3个通道(红色通道梯度、绿色通道梯度和蓝色通道梯度)和其灰度化后的灰度图像的梯度来进行特征提取，从而可以更好的突显边缘区域，为后续算法提供良好特征输入，进而达到减少线段漏检测的目的。
例如，假设彩色图像中像素(x,y)的红色通道的梯度幅值和梯度角度分别为G r(x,y)和θ r(x,y)，绿色通道的梯度幅值和梯度角度分别为G g(x,y)和θ g(x,y)，蓝色通道的梯度幅值和梯度角度分别为G b(x,y)和θ b(x,y)，灰度化后的灰度通道的梯度幅值和梯度角度分别为G gray(x,y)和θ gray(x,y)，则彩色图像梯度融合后的梯度幅值为各图像梯度幅值的最大值，即：
G fusion(x,y)＝max{G r(x,y),G g(x,y),G b(x,y),G gray(x,y)}
彩色图像梯度融合后的梯度角度为梯度幅值最大值对应的梯度角度，即：
Figure PCTCN2019098927-appb-000005
融合后的彩色图像梯度幅值综合反映了颜色和亮度的变化，融合后的彩色图像梯度角度在颜色或亮度变化区域更加一致，从而提高了边缘特征的表征能力，有效地减少线段漏检测。
在说明了图像为彩色图像时，基于图像梯度融合方法提取梯度角度值和梯度幅度值作为图像特征之后，可替换地，还可以将彩色图像进行灰度化，直接提取图像灰度化后的梯度角度值和梯度幅度值作为图像特征。
例如，假设图像灰度化后的灰度通道的梯度幅值和梯度角度分别为 G gray(x,y)和θ gray(x,y)，则直接提取梯度幅值和梯度角度分别为G gray(x,y)和θ gray(x,y)作为图像特征。
或者，当该图像不是彩色图像(例如，该图像是灰度图像)时，则直接提取图像的梯度角度值和梯度幅度值作为图像特征。
接下来，种子点确定单元1002可以基于提取的图像特征确定多个种子点。
在此，种子点确定单元1002基于提取的图像特征，通过非极大值抑制和阈值筛选确定多个种子点。
图像梯度反映了图像颜色或亮度急剧连续变化，图像梯度的非极大值抑制是沿着图像梯度方向对图像梯度幅值进行非极大值抑制，因此图像梯度非极大值抑制后的有效像素是线段边缘的候选点，这里，可将其作为连通区域生长的种子点的候选点。
例如，图像梯度的非极大值抑制公式如下所述，即如果当前像素的梯度幅值比其在梯度方向上相邻两像素的梯度幅值都大，则保留当前点梯度幅值，否则抑制当前点：
Figure PCTCN2019098927-appb-000006
其中，G NMS(x,y)表示图像梯度经非极大值抑制后的有效像素梯度，G(x,y)表示像素(x,y)的图像梯度幅值，G θ(x,y) 1和G θ(x,y) 2分别表示在图像梯度方向上与当前像素相邻的两像素的梯度幅值。即，如果当前像素(x,y)的梯度幅值G(x,y)比其在梯度方向上相邻两像素的梯度幅值(G θ(x,y) 1和G θ(x,y) 2)都大时，保留当前点梯度幅值，否则抑制当前点。
在通过非极大值抑制确定多个有效像素梯度后，接下来，通过阈值筛选来确定多个种子点，下面介绍通过阈值筛选确定多个种子点的过程。
本公开以图像的梯度幅值统计均值与0.2倍图像梯度统计方差之间的差值为阈值，将图像梯度经非极大值抑制后的有效像素梯度大于该阈值的所有有效像素点作为连通区域生长的种子点，同时将所有梯度幅值大于该图像梯度幅值统计均值和图像梯度统计方差之和的所有有效像素点作为连通区域生长的种子点，如下公式所示：
P seed(x,y)＝{{P(x,y)|G NMS(x,y)>threshold seed1}U{P(x,y)|G(x,y)>threshold seed2}}
其中，
threshold seed1＝aveMag-0.2stdMag
threshold seed2＝aveMag+stdMag
aveMag＝(1/N)∑G(x,y)
Figure PCTCN2019098927-appb-000007
其中P seed(x,y)表示种子点集合，P(x,y)表示图像像素，G(x,y)表示像素(x,y)的图像梯度幅值，N为图像像素的数目，G NMS(x,y)表示图像梯度经非极大值抑制后的有效像素梯度，aveMag表示梯度幅值统计均值，stdMag表示图像梯度统计方差。
如上所述，图像梯度经非极大值抑制后的有效像素有很多，有些属于强边缘区域，但是大部分为弱边缘或噪声区域，强边缘区域像素的梯度幅值较大，弱边缘区域像素的梯度幅值较小。经阈值筛选后的种子点绝大部分属于强边缘区域像素，这为后续的区域生长提供了良好的起始点，可在一定程度上减少虚假线段的产生，同时减少连通区域生长的计算量。
接下来，对于所述多个种子点中的每个种子点，区域生长规则确定单元1003根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则。
其中，对于多个种子点中的每个种子点，区域生长规则确定单元1003按照梯度幅值由大到小的顺序依次选取种子点作为起始生长点；对于每个起始生长点，将满足区域生长规则的其他种子点加入到该种子点的初始连通区域中作为当前连通区域。
连通区域中的像素具有相近的梯度方向以及相近的梯度幅值，本公开提出了需要同时满足梯度幅值生长规则和梯度角度生长规则的连通区域生长规则。
梯度幅值生长规则为：对每个起始生长点的邻域像素的幅值进行排序，选择幅值排在预定位次的像素作为满足梯度幅值生长规则的像素。例如，对当前生长点的八邻域像素的梯度幅值进行排序，如果某个八邻域像素的梯度幅值排在前4名，则该像素满足梯度幅值生长规则。应该认识到，以上预定位次只是一个示例，上述预定位次可以根据需要设定。
梯度角度生长规则为：将未加入到当前连通区域的像素作为待判定像素， 选择梯度角度与当前连通区域的梯度角度差值的绝对值小于预定值的待判定像素作为满足梯度角度生长规则的像素。例如，图像梯度角度生长规则与LSD直线提取算法相同，即若待判定像素梯度角度与已生长区域的梯度角度差值的绝对值小于22.5度，则其满足梯度角度生长规则。应该认识到，以上梯度角度差值只是一个示例，上述梯度角度差值可以根据需要设定。
进一步，在根据区域生长规则确定各个种子点的当前连通区域后，区域修正单元1004可以对所述当前连通区域进行区域修正，所述区域修正包括：比较连通区域内像素占其外接矩形的像素的比例和设定的占比阈值；如果连通区域内像素占其外接矩形的像素的比例小于设定的占比阈值，则通过减少图像梯度角度阈值或减少外接矩形边长重新进行区域生长。
接下来，线段拟合单元1005对各个种子点的当前连通区域内的线段进行线段拟合。即，可以首先利用图像非极大值抑制有效像素进行线段所属直线的拟合，再结合连通区域像素的图像坐标信息确定拟合线段的两个端点来确定拟合的线段。
在确定了拟合的线段之后，判断单元1006可以通过计算图像中错误线段的数目来判断线段拟合结果是否有效，即在对各个种子点的当前连通区域内的线段进行线段拟合之后，还包括如下步骤：计算图像中错误线段的数目，选择错误线段的数目小于预设阈值的线段拟合结果作为有效结果。
例如，可以通过如下方法来验证线段检测结果是否有效。对于一个由n个种子点拟合的线段，计算每个种子点到该线段的距离，并统计距离小于1.5的种子点像素的个数，然后基于该线段计算该图像中错误线段的数目(NFA)。如上所述NFA的判断方法，在这里不做重复赘述。
下面，参照图5描述根据本公开实施例的线段检测设备。图5是根据本公开实施例的线段检测设备的示意图。由于本实施例的线段检测设备的功能与在上文中参照图1描述的方法的细节相同，因此在这里为了简单起见，省略对相同内容的详细描述。
如图5所示，线段检测设备1100包括存储器1101，配置为存储计算机可读指令；以及处理器1102，配置为处理存储在所述存储器中的所述计算机可读指令，其中所述处理器处理所述计算机可读指令时执行以下功能：提取待处理图像的图像特征；基于提取的图像特征确定多个种子点；对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其 中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及对各个种子点的当前连通区域内的线段进行线段拟合。
本公开还提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序在由处理器执行时实现以下方法：提取待处理图像的图像特征，图像特征包括图像梯度角度和图像梯度幅值；基于提取的图像特征确定多个种子点；对于多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及对各个种子点的当前连通区域内的线段进行线段拟合。
本领域技术人员可以理解，本申请的各方面可以通过若干具有可专利性的种类或情况进行说明和描述，包括任何新的和有用的工序、机器、产品或物质的组合，或对他们的任何新的和有用的改进。相应地，本申请的各个方面可以完全由硬件执行、可以完全由软件(包括固件、常驻软件、微码等)执行、也可以由硬件和软件组合执行。以上硬件或软件均可被称为“数据块”、“模块”、“引擎”、“单元”、“组件”或“系统”。此外，本申请的各方面可能表现为位于一个或多个计算机可读介质中的计算机产品，该产品包括计算机可读程序编码。
本申请的各个单元可以包括微处理器、数字信号处理器(DSP，Digital Signal Processor)、专用集成电路(ASIC，Application Specific Integrated Circuit)、可编程逻辑器件(PLD，Programmable Logic Device)、中央处理器(CPU，Central Processing Unit)、图形处理器(GPU，Graphics Processing Unit)、现场可编程门阵列(FPGA，Field Programmable Gate Array)等硬件，可以通过该硬件来实现各功能块或各单元的部分或全部。
本申请使用了特定词语来描述本申请的实施例。如“一个实施例”、“一实施例”、和/或“一些实施例”意指与本申请至少一个实施例相关的某一特征、结构或特点。因此，应强调并注意的是，本说明书中在不同位置两次或多次提及的“一实施例”或“一个实施例”或“一替代性实施例”并不一定是指同一实施例。此外，本申请的一个或多个实施例中的某些特征、结构或特点可以进行适当的组合。
除非另有定义，这里使用的所有术语(包括技术和科学术语)具有与本公开所属领域的普通技术人员共同理解的相同含义。还应当理解，诸如在通常字典里定义的那些术语应当被解释为具有与它们在相关技术的上下文中的含义相一致的含义，而不应用理想化或极度形式化的意义来解释，除非这里明确地 这样定义。
以上是对本公开的说明，而不应被认为是对其的限制。尽管描述了本公开的若干示例性实施例，但本领域技术人员将容易地理解，在不背离本公开的新颖教学和优点的前提下可以对示例性实施例进行许多修改。因此，所有这些修改都意图包含在权利要求书所限定的本公开范围内。应当理解，上面是对本公开的说明，而不应被认为是限于所公开的特定实施例，并且对所公开的实施例以及其他实施例的修改意图包含在所附权利要求书的范围内。本公开由权利要求书及其等效物限定。

Claims (20)

  1. 一种线段检测方法，包括：
    提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；
    基于提取的图像特征确定多个种子点；
    对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及
    对各个种子点的当前连通区域内的线段进行线段拟合。
  2. 根据权利要求1所述的方法，其中，所述提取待处理图像的图像特征包括：
    当所述图像为彩色图像时，基于图像梯度融合方法进行图像特征提取；或者通过将所述彩色图像进行灰度化来进行图像特征提取；
    当所述图像为灰度图像时，直接进行图像特征提取。
  3. 根据权利要求2所述的方法，其中，所述图像梯度融合方法为：
    提取彩色图像的红色通道梯度、绿色通道梯度、蓝色通道梯度和灰度化后的通道梯度幅值的最大值作为融合后的彩色图像梯度幅值；以及
    提取所述融合后的彩色图像梯度幅值对应的梯度角度作为融合后的彩色图像梯度角度。
  4. 根据权利要求1所述的方法，其中，所述基于提取的图像特征确定多个种子点包括：
    基于提取的图像特征，通过非极大值抑制和阈值筛选确定多个种子点。
  5. 根据权利要求4所述的方法，其中，所述通过非极大值抑制确定多个种子点的方法如下：
    Figure PCTCN2019098927-appb-100001
    其中，G(x,y)表示像素(x,y)的图像梯度幅值，G θ(x,y) 1和G θ(x,y) 2分别表示在图像梯度方向上与当前像素相邻的两像素的梯度幅值，G NMS(x,y)表示图像梯度经非极大值抑制后的有效像素梯度。
  6. 根据权利要求4所述的方法，其中，所述阈值筛选确定多个种子点包 括：
    以图像的梯度幅值统计均值与预设倍数图像梯度统计方差之间的差值为阈值，将图像梯度经非极大值抑制后的有效像素梯度大于该阈值的所有有效像素点作为连通区域生长的种子点，同时将所有梯度幅值大于所述图像梯度幅值统计均值和图像梯度统计方差之和的所有像素点作为连通区域生长的种子点。
  7. 根据权利要求1所述的方法，其中，所述对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域包括：
    按照梯度幅值由大到小的顺序依次选取种子点作为起始生长点；
    对于每个起始生长点，将满足区域生长规则的其他种子点加入到该种子点的初始连通区域中作为当前连通区域。
  8. 根据权利要求7所述的方法，其中，所述梯度幅值生长规则为：
    对每个起始生长点的邻域像素的幅值进行排序，选择幅值排在预定位次的像素作为满足梯度幅值生长规则的像素。
  9. 根据权利要求7所述的方法，其中，所述梯度角度生长规则为：
    将未加入到当前连通区域的像素作为待判定像素，选择梯度角度与当前连通区域的梯度角度差值的绝对值小于预定值的待判定像素作为满足梯度角度生长规则的像素。
  10. 根据权利要求1所述的方法，还包括：
    在根据区域生长规则确定各个种子点的当前连通区域后，对所述当前连通区域进行区域修正，所述区域修正包括：
    比较连通区域内像素占其外接矩形的像素的比例和设定的占比阈值；
    如果连通区域内像素占其外接矩形的像素的比例小于设定的占比阈值，则通过减少图像梯度角度阈值或减少外接矩形边长重新进行区域生长。
  11. 根据权利要求1所述的方法，其中，对各个种子点的当前连通区域内的线段进行线段拟合包括：
    通过利用非极大值抑制当前连通区域内的像素来进行线段所属直线的拟合；以及
    结合当前连通区域内的像素的图像坐标信息确定所拟合的线段的两个端点。
  12. 根据权利要求11所述的方法，其中，对各个种子点的当前连通区域 内的线段进行线段拟合之后，还包括如下步骤：
    计算图像中错误线段的数目，选择所述错误线段的数目小于预设阈值的线段拟合结果作为有效结果。
  13. 根据权利要求12所述的方法，其中，通过以下方法计算图像中错误线段的数目NFA：
    Figure PCTCN2019098927-appb-100002
    其中，L表示由n个种子点拟合的线段，至少有x个种子点到拟合线段L的距离小于预定阈值，(M×N) 2表示在(M×N)大小的图像中潜在的线段数量，概率p表示种子点不是线段拟合误差点的概率。
  14. 根据权利要求12所述的方法，其中，所述预定阈值为图像中允许存在的错误线段的最大数目。
  15. 一种线段检测装置，包括：
    图像特征提取单元，配置为提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；
    种子点确定单元，配置为基于提取的图像特征确定多个种子点；
    区域生长规则确定单元，配置为对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及
    线段拟合单元，配置为对各个种子点的当前连通区域内的线段进行线段拟合。
  16. 根据权利要求15所述的装置，其中，所述提取待处理图像的图像特征包括：
    当所述图像为彩色图像时，基于图像梯度融合方法进行图像特征提取；或者通过将所述彩色图像进行灰度来进行图像特征提取；
    当所述图像为灰度图像时，直接进行图像特征提取。
  17. 根据权利要求16所述的装置，其中，所述图像梯度融合方法为：
    提取彩色图像的红色通道梯度、绿色通道梯度、蓝色通道梯度和灰度化后的通道梯度幅值的最大值作为融合后的彩色图像梯度幅值；以及
    提取所述融合后的彩色图像梯度幅值对应的梯度角度作为融合后的彩色图像梯度角度。
  18. 根据权利要求15所述的装置，其中，所述基于提取的图像特征确定多个种子点包括：
    基于提取的图像特征，通过非极大值抑制和阈值筛选确定多个种子点。
  19. 一种线段检测设备，包括：
    存储器，配置为存储计算机可读指令；以及
    处理器，配置为处理存储在所述存储器中的所述计算机可读指令，其中所述处理器处理所述计算机可读指令时执行权利要求1所述的方法。
  20. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序在由处理器执行时实现以下方法：
    提取待处理图像的图像特征，所述图像特征包括图像梯度角度和图像梯度幅值；
    基于提取的图像特征确定多个种子点；
    对于所述多个种子点中的每个种子点，根据区域生长规则确定各个种子点的当前连通区域，其中，区域生长规则需要同时满足梯度幅值生长规则和梯度角度生长规则；以及
    对各个种子点的当前连通区域内的线段进行线段拟合。
PCT/CN2019/098927 2018-08-30 2019-08-01 线段检测方法、装置、设备和计算机可读存储介质 Ceased WO2020042866A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP19850806.1A EP3846120A4 (en) 2018-08-30 2019-08-01 LINE SEGMENT DETECTION METHOD AND APPARATUS, DEVICE AND COMPUTER READABLE STORAGE MEDIUM
US16/641,830 US11276178B2 (en) 2018-08-30 2019-08-01 Line segment detection method, apparatus, device, and computer-readable storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811005060.7A CN110874844A (zh) 2018-08-30 2018-08-30 线段检测方法、装置和设备
CN201811005060.7 2018-08-30

Publications (1)

Publication Number Publication Date
WO2020042866A1 true WO2020042866A1 (zh) 2020-03-05

Family

ID=69643849

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/098927 Ceased WO2020042866A1 (zh) 2018-08-30 2019-08-01 线段检测方法、装置、设备和计算机可读存储介质

Country Status (4)

Country Link
US (1) US11276178B2 (zh)
EP (1) EP3846120A4 (zh)
CN (1) CN110874844A (zh)
WO (1) WO2020042866A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112580447A (zh) * 2020-12-04 2021-03-30 国网福建省电力有限公司泉州供电公司 一种基于边缘二阶统计与融合的电力线检测方法
CN113221926A (zh) * 2021-06-23 2021-08-06 华南师范大学 一种基于角点优化的线段提取方法
CN114005130A (zh) * 2021-10-09 2022-02-01 南京信息工程大学 一种基于lsd线段检测的文档图像三角图检测方法
CN115482206A (zh) * 2022-09-01 2022-12-16 河北工业大学 一种超大型钢筋网片整体尺寸检测方法
CN115861317A (zh) * 2023-02-27 2023-03-28 深圳市伟利达精密塑胶模具有限公司 基于机器视觉的塑胶模具生产缺陷检测方法

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113763379B (zh) * 2021-09-23 2023-07-18 成都唐源电气股份有限公司 吊弦断股检测方法、装置、计算机设备及存储介质
CN116958166A (zh) * 2022-04-20 2023-10-27 武汉联影医疗科技有限公司 图像分割方法、设备、存储介质和程序产品
CN114862890B (zh) * 2022-05-26 2025-04-11 南京信息工程大学 一种基于改进lsd的磁极倒转图信息抽取方法
CN114878595B (zh) * 2022-07-08 2022-09-20 山东蓝彩天下教育科技有限公司 一种书本印刷质量检测方法
CN115545094A (zh) * 2022-09-13 2022-12-30 国能信控互联技术有限公司 基于区域生长的火电机组稳定工况识别方法、终端及介质
CN115330791A (zh) * 2022-10-13 2022-11-11 江苏东晨机械科技有限公司 一种零件毛刺检测方法
CN115861291B (zh) * 2023-01-06 2023-05-09 曲阜天博汽车电器有限公司 基于机器视觉的贴片电路板生产缺陷检测方法
CN116051543B (zh) * 2023-03-06 2023-06-16 山东锦霖钢材加工有限公司 一种用于钢材剥皮的缺陷识别方法
CN116485904A (zh) * 2023-03-26 2023-07-25 重庆大学 基于图像梯度阈值计算的改进移动机器人EDLines线段检测方法
CN116612297B (zh) * 2023-05-08 2025-07-11 华南师范大学 一种融合轮廓特征的图像线段提取方法
CN116228764B (zh) * 2023-05-08 2023-07-18 聊城市东昌府区妇幼保健院 一种新生儿疾病筛查血片采集质量检测方法及系统
CN116664552B (zh) * 2023-07-24 2023-09-29 凯尔菱电（山东）电梯有限公司 基于视频图像分析的电梯门异常检测方法
CN118229574B (zh) * 2024-05-21 2024-08-06 济宁景泽信息科技有限公司 一种给排水设计图数字化增强方法
CN119941667A (zh) * 2025-01-03 2025-05-06 华南农业大学 基于机器视觉的滑块尺寸测量方法、装置、设备及介质
CN119648689B (zh) * 2025-02-12 2025-05-02 西安宝莱特光电器件有限公司 基于机器视觉的oled显示屏质量检测方法
CN120088216B (zh) * 2025-02-12 2025-08-19 吴江市雄狮漂染有限公司 一种布料染色着色均匀度视觉检测方法及系统
CN120912892B (zh) * 2025-10-11 2025-12-02 南京邮电大学 高分辨率sar卫星影像线段提取方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009150545A2 (en) * 2008-06-09 2009-12-17 Uti Limited Partnership Methods, systems, and computer program products for gpu-based point radiation for interactive volume sculpting and segmentation
CN102496136A (zh) * 2011-12-08 2012-06-13 中国人民解放军国防科学技术大学 基于图形处理单元的相位编组直线提取方法
CN102750703A (zh) * 2012-06-27 2012-10-24 北京航空航天大学 一种基于直线分割的遥感图像机场自动检测方法

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20010033680A1 (en) * 1995-10-26 2001-10-25 Bankman Isaac N. Method and system for detecting small structures in images
US5991512A (en) * 1997-07-31 1999-11-23 Hewlett-Packard Company Ink relocation for color halftones
DE10012875B4 (de) * 2000-03-16 2004-04-01 Infineon Technologies Ag Mobilfunkempfänger
US6778699B1 (en) * 2000-03-27 2004-08-17 Eastman Kodak Company Method of determining vanishing point location from an image
US7983446B2 (en) * 2003-07-18 2011-07-19 Lockheed Martin Corporation Method and apparatus for automatic object identification
US7679620B2 (en) * 2005-07-28 2010-03-16 Microsoft Corp. Image processing using saltating samples
CN103473762B (zh) * 2013-08-29 2016-12-28 奇瑞汽车股份有限公司 一种车道线检测方法和装置
CN104680506A (zh) * 2013-11-28 2015-06-03 方正国际软件(北京)有限公司 一种分方向边界线检测方法及系统
CN103927743B (zh) * 2014-03-27 2017-04-05 中国科学院长春光学精密机械与物理研究所 一种遥感成像中人造目标的检测方法
CN104657490A (zh) * 2015-03-05 2015-05-27 四川智羽软件有限公司 一种信息检索方法
US10013608B2 (en) * 2015-07-17 2018-07-03 Tata Consultancy Services Limited Method and system for facilitating real time detection of linear infrastructural objects by aerial imagery
CN105069786B (zh) 2015-07-31 2018-05-08 小米科技有限责任公司 直线检测方法及装置
CN105261031B (zh) * 2015-11-26 2019-01-15 四川汇源光通信有限公司 基于梯度计算的直线检测方法及装置
EP3264335A1 (en) * 2016-06-28 2018-01-03 Services Pétroliers Schlumberger Automatic dip picking in borehole images
CN107273903B (zh) * 2017-05-22 2020-06-16 哈尔滨工程大学 基于lsd改进的uuv近海面可见光图像海天线提取方法
CN107463939B (zh) 2017-06-27 2020-06-09 浙江工业大学 一种图像关键直线检测方法
CN108416789A (zh) * 2018-06-04 2018-08-17 武汉斑马快跑科技有限公司 图像边缘检测方法及系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009150545A2 (en) * 2008-06-09 2009-12-17 Uti Limited Partnership Methods, systems, and computer program products for gpu-based point radiation for interactive volume sculpting and segmentation
CN102496136A (zh) * 2011-12-08 2012-06-13 中国人民解放军国防科学技术大学 基于图形处理单元的相位编组直线提取方法
CN102750703A (zh) * 2012-06-27 2012-10-24 北京航空航天大学 一种基于直线分割的遥感图像机场自动检测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
See also references of EP3846120A4 *
SHI, . JINGYE ET AL.: "Mentation Based on Modified Region Growing Algorithm", OPTICAL TECHNIQUE, vol. 43, no. 4, 31 July 2017 (2017-07-31), pages 382 - 383, XP055690121, ISSN: 1002-1582 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112580447A (zh) * 2020-12-04 2021-03-30 国网福建省电力有限公司泉州供电公司 一种基于边缘二阶统计与融合的电力线检测方法
CN112580447B (zh) * 2020-12-04 2023-08-04 国网福建省电力有限公司泉州供电公司 一种基于边缘二阶统计与融合的电力线检测方法
CN113221926A (zh) * 2021-06-23 2021-08-06 华南师范大学 一种基于角点优化的线段提取方法
CN114005130A (zh) * 2021-10-09 2022-02-01 南京信息工程大学 一种基于lsd线段检测的文档图像三角图检测方法
CN115482206A (zh) * 2022-09-01 2022-12-16 河北工业大学 一种超大型钢筋网片整体尺寸检测方法
CN115861317A (zh) * 2023-02-27 2023-03-28 深圳市伟利达精密塑胶模具有限公司 基于机器视觉的塑胶模具生产缺陷检测方法
CN115861317B (zh) * 2023-02-27 2023-04-28 深圳市伟利达精密塑胶模具有限公司 基于机器视觉的塑胶模具生产缺陷检测方法

Also Published As

Publication number Publication date
CN110874844A (zh) 2020-03-10
EP3846120A1 (en) 2021-07-07
US11276178B2 (en) 2022-03-15
US20210150729A1 (en) 2021-05-20
EP3846120A4 (en) 2022-05-18

Similar Documents

Publication Publication Date Title
WO2020042866A1 (zh) 线段检测方法、装置、设备和计算机可读存储介质
US12394036B2 (en) Information processing device, information processing method, and storage medium
CN108629343B (zh) 一种基于边缘检测和改进Harris角点检测的车牌定位方法和系统
CN106960451B (zh) 一种提升图像弱纹理区域特征点数量的方法
CN104091324B (zh) 一种基于连通域分割的棋盘格图像快速特征匹配算法
JP6445775B2 (ja) 画像処理装置、画像処理方法
CN113095385B (zh) 一种基于全局和局部特征描述的多模图像匹配方法
CN105320927B (zh) 车道线检测方法及系统
CN103679737B (zh) 基于多通道信息选择的彩色图像边缘检测方法
CN112749645B (zh) 基于监控视频的服装颜色检测方法、装置和设备
CN112101370A (zh) 一种纯色背景图像自动抠像算法、计算机可读存储介质及设备
CN110009653A (zh) 基于灰度阈值分割法增加边缘识别点准确图像识别的方法
CN112396050B (zh) 图像的处理方法、设备以及存储介质
CN108389215B (zh) 一种边缘检测方法、装置、计算机存储介质及终端
CN104778710B (zh) 一种基于量子理论的形态学图像边缘检测方法
US20150131853A1 (en) Stereo matching system and method for generating disparity map using same
CN105913451B (zh) 一种基于图模型的自然图像超像素分割方法
CN114863492A (zh) 一种低质量指纹图像的修复方法及修复装置
CN107292238A (zh) 基于空间拓扑关系约束和消失点的建筑物轮廓识别方法
CN104463896B (zh) 基于核相似区分布特性的图像角点检测方法及系统
CN113221926B (zh) 一种基于角点优化的线段提取方法
CN112001927B (zh) 基于线基元几何空间关系连接的电力线检测方法
CN107103607B (zh) 一种基于种子点自动选取的全局最优化图像分割方法
CN108717699A (zh) 一种基于连续最小割的超声波图像分割方法
CN112258561B (zh) 针对图像拼接的匹配点获取方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19850806

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019850806

Country of ref document: EP

Effective date: 20210330