JPH03295355A - Picture read method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image reading method in which images such as characters and figures are input by manual feeding operation.

[Prior Art] A conventional reading method using two one-dimensional image sensors in an image reading device using manual feeding operation is disclosed in Japanese Patent Laid-Open No. 61-78274. This method involves arranging a first sensor and a second sensor at regular intervals in the hand-feeding direction, and calculating the exclusive OR of the image read by the first sensor at a certain position and the image read by the second sensor. When the degree of coincidence exceeds a certain value, it is assumed that the second sensor has come to the reading position of the first sensor, and the speeds of the first and second sensors are detected from the number of reading scans during that time, and the image Reading control is in progress.

[Problem to be solved by the invention]

However, the method does not assume that the input image contains noise or the like. Therefore, if noise is present in the input image, the degree of coincidence will be low if the degree of coincidence is determined by the exclusive OR of the images obtained from the two sensors.

If the degree of coincidence is compared and the input images are associated in such a state, even if a corresponding image actually exists, it may not be possible to find it because the degree of coincidence is low. As a result, the search continues until the feature point is found, making it impossible to correct image distortion. Or, there is a possibility that it will be connected to a feature point that appeared later, making it impossible to perform accurate distortion correction.

Therefore, the present invention is intended to solve the above-mentioned problems, and its purpose is to accurately calculate features even if noise is included in the input image, and to detect the speed of image matching. An object of the present invention is to provide an image reading method that obtains images without distortion.

[Means for Solving the Problems] The present invention has at least two sensors, a first sensor and a second sensor, each of which is a one-dimensional image sensor that reads characters and graphics, and the first sensor that moves forward in the hand-feeding direction. In a method of reading a W image while correcting image distortion by detecting the time taken for an image read by one sensor at a certain position to be read by the second sensor proceeding later, the white image or black image in a line of the input image is detected. It determines whether a short line in the image is noise based on the image trend, and if it is noise, it is absorbed into the surrounding images and its features are calculated.The features are compared to detect the speed and image distortion is corrected. It is characterized by doing.

[Example] Hereinafter, the present invention will be explained in detail based on examples.

FIG. 1 is a diagram illustrating a method of the present invention for determining noise in an input image (line) and detecting accurate features of the image. FIG. 2 shows a method of comparing the features and calculating the degree of coincidence in terms of force. FIG. 3 shows an image reading device having two sensors arranged parallel to the manual feeding direction. FIG. 4 shows images obtained from two sensors input by the image reading device.

The image reading method of the present invention will be explained below.

The image reading device used in the present invention has a manual feed direction 308.
Two one-dimensional image sensors (3
02 and 303), and the reflected light of an image 307 is formed on the two one-dimensional image sensors by a lens 306. Further, the image 307 is always illuminated by a light source (not shown) such as an LED.

When the image input device is manually moved in the manual feed direction 308 to perform manual image capture, two one-dimensional image sensors (302 and 3
03), images of the areas shown in 304 and 305 are manually input. The -dimensional image sensor (302 and 30
3) are arranged in parallel with an interval d30114, and since the lens 306 is set to project images one-to-one, the interval 309 between the reading positions 304 and 305 is the same as the interval d301. Further, the manual feeding direction of the image input device is the direction indicated by 308, and the lens 306
Since the image information at position W305 is input to the sensor 303 and the image information at position 304 is input to the sensor 302, the distance d301 is smaller than the image input to the sensor 303 (hereinafter referred to as the first sensor). An image of the same point is input to the sensor 302 (hereinafter referred to as the second sensor) after a travel time delay of .

Next, a method of detecting feature points of an image input by hand-feeding operation using the image input device and correcting distortion will be described. When the image shown in FIG. 4(a) written on the manuscript is input by manual feeding operation using the image input device shown in FIG. 3, the first sensor 303 inputs the image shown in FIG. The image shown in FIG.
The image shown in Figure (C) is obtained. Note that the two images shown in FIGS. 4(b) and 4(c) have the time axis 401 in the horizontal direction.
Images located at the same time on the time axis 401 are input at the same time. Therefore, by detecting the time delay between the images at the same point in the two images, the moving speed of the image input device during this time can be determined from the distance d between the first sensor and the second sensor, and an image without distortion can be obtained. becomes possible. Further, the correspondence between feature points of images may be determined by calculating the exclusive OR of two images as in JP-A-61-78274, but in the present invention, as shown in FIG. The boundary points between the white image and the black image are associated with each other, and the degree of coincidence is calculated based on the amount of deviation, and speed correction is performed.

In the present invention, it is determined whether noise is present in the input image before comparing the degree of matching. As a method of judgment, first focus on the overall flow of the image. Figure 1 (a
) 101, the image is 111.112 from the left
．． It consists of five groups: 113, 114, and 115. Here, we will focus on the flow of the lengths of the five groups of regions 111 to 115. In this case, the length of the region 112 is extremely short compared to the lengths of the other regions, which disturbs the flow of the length of the entire region.

Therefore, the area 112 is likely to be noise. Therefore, in the present invention, such areas where noise is likely to occur are absorbed into the surrounding areas, that is, the areas 111 and 113. By doing so, Figure 1 (a) 10
The image 1 becomes an image 102, which is a noise-free image. In this way, accurate image comparisons are possible by creating noise-free image features. for example,
The image 101 (201) with noise and the image 102 (203) without noise are combined into the correct image 202.
(or 204) with respect to the boundaries of each area of the image (
211-2] 4 and 215-216 and 241-242 and 243-244), the results are shown in FIGS. 2(a) and 2(b). This shows how much the degree of matching differs depending on whether noise 222 is present or not. This means that the presence of noise 222, which is only 5% of the total, reduces the degree of matching of the images to 45%.

The influence of noise on the degree of matching differs depending on the method of feature comparison, but the influence of noise is significant no matter how the features are compared. The areas indicated by 231 to 233 and 261 in FIG. 2 mean matching areas.

Furthermore, the present invention determines whether or not it is noise by focusing on the flow of the entire image.

Therefore, the area 1 determined to be noise in the image 101
Even if the area is smaller than 12, if each area is small and does not disrupt the overall flow, as in image 103, it is not determined to be noise. In image 103, only area 137 was determined to be noise, and this area is similar to areas 136 and 138.
It is absorbed into the image 104 and becomes an area ]56. In this way, since it is determined whether or not it is noise by focusing on the overall flow, noise can be accurately removed without damaging the features of the image, and accurate features can be extracted.

As described above, a feature point is extracted from the image input from the first sensor, an image corresponding to the feature point is searched from the image input from the second sensor, and the image input device By detecting the moving speed of the camera and correcting image distortion, it is possible to obtain clear images. Further, in the present invention, noise is determined by focusing on the flow of the size of each region of the entire image. As a result, even if it is not actually noise, because the area is small compared to other areas, it may be determined to be noise and absorbed by surrounding areas. However, since such images also absorb the corresponding images in the same way, the characteristics of the corresponding images match.

As described above, according to the present invention, attention is paid to the overall flow of the length of the white image area and black image area of the input image, and in the case of an image with a small area that disturbs the flow of the length of the area, is determined to be noise and is absorbed into the surrounding images to extract the features, making it possible to compare accurate features unaffected by noise. As a result, it becomes possible to accurately detect the speed of the image input device, and it becomes possible to obtain a clear image without distortion.

〔Effect of the invention〕

As described above, according to the present invention, attention is paid to the flow of the white image or black image area in the input line, and if the area is extremely small compared to the overall flow, it is determined that it is noise. 7': Since good features are compared, the degree of matching may be low due to the presence of noise, or the images may be judged as different images even though they are the same. This eliminates the need for accurate feature comparisons.

As a result, the correspondence between feature points becomes accurate, and it becomes possible to obtain a clear image without distortion.

The image reading method of the present invention eliminates the need for an encoder for positioning, etc., making it possible to further downsize a small image input device and provide an image input device that is extremely easy to use.

[Brief explanation of drawings]

FIG. 1 is a detailed explanatory diagram of the present invention. FIG. 2 is an explanatory diagram of matching degree detection by comparing features. FIG. 3 is a configuration diagram of an image reading device used in the present invention. FIG. 4 is an explanatory diagram of an image input by the image reading device. 101... Input image before noise removal 102... Input image after noise removal 103... Input image before noise removal 104... Input image after noise removal 301... Sensor interval 302...・・Second sensor 303・・・・First sensor 304 ・ 305 ・ 306 ・ 307 ・ 401 ・・Second sensor reading position・First sensor reading position・Lens・Reading original image・More than time axis

Claims (1)

[Claims] It has at least two sensors, a first sensor and a second sensor, which are one-dimensional image sensors that read characters and figures, and the image read at a certain position by the first sensor, which moves forward in the hand-feeding direction, is later read. In the method of reading an image while detecting the time until the second sensor reads the image and correcting the image distortion, it is determined whether a short line in the image is noise based on the tendency of white images or black images among the lines of the input image. An image reading method is characterized in that, in the case of noise, the characteristics are calculated by absorbing it into the surrounding image, and then the characteristics are compared to detect the speed and correct the distortion of the image.