BRPI0802865A2 - image file generation process with automated optical distortion correction method - Google Patents

Abstract

IMAGE FILES GENERATION PROCESS WITH AUTOMATED OPTICAL DISTORTION CORRECTION METHOD. The present invention relates to a process for generating digital files from an image capture device which includes an automatic method for orthogonalizing perspective and removing optical distortions arising in the digital image, making it possible to detect the original shape and dimension characteristics of flat objects arranged on a template for later use in CAD software. The process of generating digital files from an image capture device, which includes an automatic method for correcting optical distortion, proposed by the invention has special application in the textile industry for automating the production of garment cutting molds, although it may be applied to any other economic activity in which the method is appropriate. To make flat object image correction possible, the digital image is captured along with a correction template. Subsequently, the image is then analyzed and processed by the method of the invention, yielding two crucial results that allow accurate detection of the flat object: - parameters on the image scaling to obtain the actual dimensions of the objects; - A new digital image, representing the image obtained as if the device were positioned orthogonally to the focused objects, as well as free of optical distortions that modify the shape of these objects. From the above results, using pattern recognition techniques in digital images, we extract the information from the photographed flat objects: real shape and dimensions. The process with the automated correction method was conceived aiming at its autonomy and high productivity, eliminating the specialized human labor, previously necessary in the operation of the old device used in the operation - the digitizing tablet.

Description

"PROCESS OF GENERATING IMAGE FILES WITH AUTOMATED OPTICAL DISTORTION CORRECTION METHOD"

The present invention relates to a process for generating digital files from an image capture device, which includes an automatic method for orthogonalizing perspective and removing optical distortions arising in the digital image, making it possible to detect the original shape and dimension characteristics of flat objects arranged on a template for later use in CAD software.

The process of generating digital files from an image capture device, which includes an automatic method for correcting optical distortion, proposed by the invention has special application in the textile industry for automating the production of garment cutting molds, although it may be applied to any other economic activity in which the method is appropriate.

To make flat object image correction possible, the digital image is captured along with a correction template. Subsequently, the image is then analyzed and processed by the method of the invention, yielding two crucial results that allow accurate detection of the flat object:

- parameters on image scaling to obtain the actual dimensions of the objects;

- a new digital image, representing the image obtained as if the device were positioned orthogonally to the focused objects, as well as free of optical distortions that modify the shape of these objects.

From the above results, using pattern recognition techniques in digital images, we extract the information from the photographed flat objects: real shape and dimensions.

The process with the automated correction method was conceived aiming at its autonomy and high productivity, eliminating the specialized human labor, previously necessary in the operation of the old device used in the operation - the digitizing tablet.

TECHNICAL STATE

Some methods for correcting distortion of digitized images are known, but they cannot be considered as automated methods, since they always depend, at some point, on human intervention. The main features and limitations of these known methods are described in detail below:

PATENT EP 0 324 561 A2

It deals with a method for correcting optical distortions using a grid under the object. The camera is in a fixed position, so it does not deal with change of perspective as the device is always orthogonal to the object. The methodology presents precision problems due to technological constraints of the time (1989), such as the processing capacity of digital computers, leading to an extreme simplification of the technique in order to make it viable. This causes poor accuracy problems when detecting objects.

The differences in the state of the art method from that of the invention are as follows:

- does not specify in detail the steps of the initial calibration to remove optical distortion using the grid added to the base under the object, nor how to detect the grid, and can not be considered as automatic, because it needs the help of the user. for these functions;

- the camera is placed in a fixed position, adapted to a type of mechanical arm that keeps it in orthogonal position of the image, not allowing movement to photograph in perspective, which compromises the practicality and operability of the method;

- the grid is only for initial calibration of the image acquisition device, without any processing for image planning;

- The requirement for camera fixation equipment as well as a self-illuminating base in a horizontal position makes the method less practical and versatile, making your equipment occupy a large floor area as well as complications. logistics and ultimately becoming more costly.

PATENT OF 4,331,188 A1

This is a method that employs a scanner-like reading device to read the image. The reading is simultaneous of the object and the template, and the template consists of several concentric circles. The process comprises the following steps:

- the first step is to locate the vector data of the concentric circles correction template), discarding the other vector data of the object (s) being scanned;

- the second step is to segment the entire image into various regions (triangles);

- the third step is to apply a coordinate transformation to each of the triangular regions, thereby mapping each of the coordinates of points inserted within the triangle to a new coordinate;

- if all the transforms are correct, all points of the image are mapped to a new image, which is supposed to be corrected for mechanical, optical distortion, electrical signal processing errors, etc.

The differences in the state of the art method from that of the invention are as follows:

- uses a cost-effective scanner to read the image, which occupies a large floor area where every m2 is precious;

- the template is composed of groups of concentric circles;

- works on template detection using vectorized image data;

- each of the groups of concentric circles of the template must be recognized, otherwise it is impossible to correct the image;

- Template identification is more complex and quite ineffective as it uses vector data and requires recognition of all groups of concentric circles and may require rework until an adequate image is detected for detection;

- is based on the correction of each of the triangular regions, where there will be as many different transformations as there are number of triangles;

- is more susceptible to errors as it requires that each of the transformations applied to each of the triangular regions, independent of each other, be correct so that the result is not distorted;

- It is difficult to use the scanner for simultaneous correction of various objects, which directly influences the total scanning time; - all image distortions (optical, mechanical, etc.) are corrected in one step, represented by the transformation to which the triangular region is subjected.

PATENT JP 2002366947 A

This is a method that uses a grid template positioned in front of the object to be detected. Using the mouse, the user must point in the software the 4 extreme points of the grid, the number of squares (vertical and horizontal), as well as their dimensions (spacing between grid lines). The system automatically plots all lines that make up the grid, the user must then click on each vertex of the grid, and the system replaces the lines with curves that represent the distorted grid as a result of using optical lenses. Therefore, there is no automated detection of image characteristics, the user points these characteristics so that the system calibrates at the end of the process, leading to too much human interference in the system, making it more susceptible to passive errors. need more effort on the part of the operator. There is no specific calibration for the device that will record the image (camera) for the analysis and calculation of the intrinsic characteristics of the device that give rise to optical distortions. Appointing the vertices is done manually, and at the end, the system calculates the distortions based on the information provided by the user, correcting the image. As the grid passes over the object, the system still needs to do some reconstitution of the object's edges so that it can be extracted properly.

The differences in the state of the art method from that of the invention are as follows:

- is a non-automated method, requiring the user to inform the system the correct location of the grid vertices; The method is not able to find these characteristics automatically, evidencing its lower artificial intelligence;

- the distortions are corrected through a set of curves that represent the grid lines, which are formed with the help of the user, which points the grid vertices, which demands greater operational effort to which the operator is subjugated. , making the system also more susceptible to human error;

- Perspective distortion is corrected by pointing to the four corners of the grid, whereas optical distortions are corrected at the end of the curve assembly process on the grid lines in the image.

- The calibration grid overlaps the object, requiring the object to be rebuilt at the end of the process.

PATENT JP 2004288148 A

This is a method that uses a grid template positioned behind the object to be detected. The calibration of the optical distortion removal device (camera) employs a methodology which comprises: obtaining an image of the grid only with no object to be photographed; recording the camera parameters that were detected by the method and after this first photo, another photograph is obtained with the presence of the object and thus, using the information about the previously acquired parameters, the second image acquired by the device is corrected, removing the optical distortions. Through the mouse, four or more points of the grid must be pointed in the software, so that the perspective correction of the image can be done and with this, finally, the final result is obtained.

The differences in the state of the art method from that of the invention are as follows:

- is a non-automated method, as it requires user intervention, informing the system of the correct location of at least four grid vertices; The method is not able to find such vertices automatically, evidencing their lower artificial intelligence;

- for calibration of the image capture device to remove optical distortion at least two images are required; a first photograph of the calibration template only, which is used for the calculation of the camera's internal parameters, and a second photograph with the object and grid, so that it corrects the second image and the object can be extracted in measurements. real; The obligation mentioned above is quite undesirable, because the lower value digital cameras and more widespread because they offer good image quality and low cost make several automatic adjustments even when manually configured, which makes their use unfeasible, otherwise the image will be corrected. , the objects still have some optical distortion.

BR PI 0405039-8

This patent application was filed by the same applicant under the title "Digital Image Correction System by Template" and deals with "a system, applied in the digitization of molds, which corrects dimensional errors of the image captured by digital camera by capturing image of a reference template. " The method of correcting the scanned template image on the template cites the following steps:

1. location of inflection points (corners) throughout the image;

1.1 detection of inflection points;

1.2 refinement of the position of the inflection points;

1.3 grouping or clustering of the inflection points found;

2. calculation of a homography matrix;

3. removal of non-standard pixels;

4. calculation of a new homography matrix and perspective correction;

5. correction of lens distortion.

However, the method is insufficiently described, without the minimum detail required, not allowing even a person skilled in the art to reproduce it.

INVENTION SOLUTION

The object of the present invention is a process of generating digital files from an image capture device which includes an automatic method for correcting optical distortions and which does not require any assistance from the operator. This means that from the acquisition of the digital image by the capture device and its insertion in the computer, the simple execution of the method guarantees a positive result (image corrected for optical distortion and perspective), or totally invalid (it is not possible to correct the image). ) without any interference from the operator.

The process of the invention results in the following advantages over all those prior art members:

- no human intervention;

- shorter total execution time, because it is directly and directly linked to the performance of the digital computer that is used by the method;

- passivity nullifies the occurrence of errors by the operator, since the system does not need its interference, whatever;

- High productivity, speed, full automation, reliability and accuracy of results.

Figure 1 illustrates a frame (1) on whose edges is positioned a template (2), which also delimits its useful area. The template (2) comprises orthogonal alignments (3) of alternating semicircles (4) interconnected by their vertices (5). Table (1) can have different template formats (size and number of semicircles) to meet the needs of special customers. The jig (2) with this distribution results in greater practicality, as it does not require the operator to be careful not to hide vertices (5) scattered throughout the useful area during the fixation of the objects and, therefore, the number and positioning of the objects does not influence in the execution of the method. By object is meant molds for cutting garments or any other flat figure.

The operator of the image capture device, preferably a digital camera, has free movement in front of the frame (1) to which the objects are attached, and can choose the capture angle that suits him or merely, position himself in front, but without worry about placing yourself in a position that is precisely orthogonal to the plane of objects. This proves the effectiveness of the method of the invention in correcting the perspective of the image by orthogonalizing it in a fully automated and humane free manner, a feature not achieved by prior art image correction methods.

Because they do not need and do not provide any type of equipment to fix the capture device, there is a lower acquisition cost than those proposed in a fixed position. However, regardless of having the flexibility of free movement, the user can also choose to fix the device if it is convenient, whether the objects are placed horizontally or vertically.

Preferably, the frame (1) may be magnetic, the objects to be captured being affixed with magnets and the frame being upright, on whose edges the template (2) is printed. This, added to the free movement of the operator in front of the frame, confers an important feature: that of minimizing the useful area required for its use, since it is not necessary more than a simple wall where the frame is fixed. It is also important to emphasize that the large number of objects that can be attached to the frame (1) delimited by the template (2) do not influence the final result of the method execution and, with this, the method also adds high productivity. The method allows the entire useful area of the frame (1) to be occupied by several objects, which is not possible by state of the art methods, which are limited to only one captured image.

GENERAL DESCRIPTION OF THE INVENTION PROCESS

The object of the present invention is a process for generating digital files from an image capture device, which includes an automatic method for correcting optical distortions arising from image capture: oblique perspective to the plane of the objects in focus and the radial distortions produced by the lens system. To solve both problems, the process employs an acquired image correction template and follows the steps below.

Image capture

The operator captures the image by pinning the objects inside the jigged frame and generating a digital image of the objects and jig through a capture device such as a digital camera.

Image Insertion

Inserting the scanned object image file with optical distortion into the computer.

Feedback Location

Includes the search for the vertices (5) of the semicircles (4) that make up the template (2). By finding a percentage of them and therefore not requiring their full recognition, the method is in possession of all the data needed to begin image regularization.

Jig Vertex Classification

The four vertices (6) on the sides of the rectangle formed by the template (2) are classified into four groups, each group corresponding to one side of the rectangle.

Colinearize the Vertices of Each Side of the Template

The method interacts to collinearize the vertices (6) on each side of the template (2) and thereby removes optical (radial) distortion from the captured image. At this stage, the method is in possession of an image without optical distortion.

Optionally, template (2) may employ other aligned geometric shapes rather than semicircles (figure 1), for example triangles (figure 2) or rectangles (figure 3), provided that their vertices on the same side of the rectangle forming the jigs are collinear, requiring little adaptation of the method for its execution, and not constituting a different result from the one proposed by the present invention.

Image Orthogonalization

Having an image without optical distortion remains only the orthogonalization of the image. Once the four sides (3) that make up the rectangle formed by the template (2) are known, the four vertices (6) of this rectangle are also known by the intersection of their sides and, through the projective geometry, making use of a homographic matrix. , it is known that it is possible to navigate from oblique perspective to orthogonalized projection, and vice versa.

DETAILED DESCRIPTION OF THE PROCESS

The process of generating digital files from an image capture device, which includes an automatic method for correcting optical distortions arising from image capture, object of the present invention, is now described in detail based on the following figures. attached, listed below:

Figure 1 - front view of the jig used in the correction method;

Figure 2 - front view of a jig option;

Figure 3 - front view of another jig option;

Figure 4 - flowchart of the digital image generation process;

Figure 5 - Flowchart of the distortion correction method in the digital image;

Figure 6 - Flowchart of homographic matrix calculation.

Figure 4 illustrates the flowchart of the digital image file generation process from objects arranged on a template comprising the following operations:

- capture the image (A): - fix the objects inside the frame with molded edge (A1)

- capture image of objects and template (A2)

- insertion of the scanned image into the computer (B)

- image processing and execution of distortion correction method of captured scanned images (C)

- test to verify image has been corrected (D)

- final image correction (E) - use of tools to add and / or remove scanned features (E1)

- CAD format file generation (E2) Figure 5 details the method of correcting the distortion of the digitized image indicated in step (C) of the flowchart of the digital imaging process employing illustrated in Figure 4. The method is initiated by data entry (captured scanned image - B) to the software responsible for its execution. Then the method performs a general procedure (C1) that starts by locating the template vertices through a corner detection algorithm (C11), so that it locates, throughout the image, vertices or corners of the objects present in the image, among them, the correction template itself. The distinction between template vertices and other image object vertices is made using parameters that are provided to the detection algorithm, previously calculated and incorporated into the system. However, this distinction may not be sufficient, as there may be corners of other objects that are very similar to those of the template, which explains the need for the next step.

Found the template vertices (C11), the method uses this data to make a first estimate of the homographic matrix (C12) which, coming from the projective geometry, is nothing more than a coordinate transformation matrix, as will be detailed later with base on figure 6.

As expected, this first matrix (C12) is neither accurate nor sufficient, since at the moment the image also has optical distortions. However, this first estimate of the matrix (C12) plays a key role: since the corner location algorithm (C11) cannot guarantee that only the corners of the correction template are pointed, and since the process does not allow the user to In order to automate the method, calculating a homographic matrix at this moment aims to enable the deletion of the corners found that do not belong to the template (C13). This is possible because if the method is capable of estimating a homographic matrix with sufficient margin of error, it will also know which portions of the image correspond to the template and thus can simply discard the corners detected in the rest of the image. Otherwise, the image is not sufficiently resolution or permanently damaged, such as a picture that has been blown out of focus. If so, the method interrupts its execution, since the image is not considered adequate to detect the characteristic of the objects present in it, since it would produce results below the desired quality and precision.

Discarding the corners that do not refer to the template, and even though only a few remain, the method is now able to calculate an improved estimate of the homographic matrix (C14).

With an improved homographic matrix (C14) and even optical distortions still present, the method applies this matrix to a copy of the detected vertices (C15), aiming to equate the corrected perspective vertices with data that would virtually be the location. the exact vertices of the template. This is possible because the method knows in advance the number of vertices present in the template as well as the spacing between them.

Thus, the test (C16) is performed which determines whether a minimum number of points is successfully matched. If so, the method continues executing, otherwise it is interrupted. This approach has been adopted to ensure the high accuracy of the results because if the method needs to be stopped it means that the image is not of sufficient quality locally, such as only a portion of the image that has had its focus impaired, which can happen for several handling factors as well as the very quality of capture devices such as amateur digital cameras.

Finally, the method is able to proceed to the correction of optical distortion (C2). Matched template vertices are the fundamental input data for this step, which then interacts to look for the coefficients of the equation variables that, when applied to the image, incur a coordinate transformation to each of its pixels, constituting a new image. , free from radial optical distortion. According to the method proposed by Tsai [Tsai87], the radial distortion equation can be approximated using only the first term of an infinite series, r '= r + kr3, where r'is the distorted radius, r is the original radius, ek is the radial distortion coefficient. To find the coefficient, we use a least squares algorithm (C21), such as the Levenberg-Marquardt nonlinear function minimizing algorithm, which automatically computes the distortion coefficient based on the information of the vertices of each of the 4 relative lines. correction template, and finally enabling image correction for radial optical distortion (C22).

For orthogonalization of the image perspective (C3) the method performs a new general procedure (C1 = C31) from the first step (C11) to before the step that removes radial distortion (C2) as it is already working with an image. corrected for optical distortion. Thus, once again the corner detection algorithm (C11) is used, an initial homographic matrix (C12) is estimated to exclude non-template vertices or corners (C13), and finally, it is calculated a definitive homographic matrix (C14), applying to the optical distortion corrected image (C15), generating the final result proposed by the orthogonalized image method (C32).

Figure 6 illustrates the procedure for calculating the homographic matrix (C12) that is performed using the Hough Transform (C121), whose generated data make it possible to define a line that best represents each of the 4 corners of the rectangle formed by the correction template. This is done as follows: For all vertices found by the corner detection algorithm (C11), the algorithm creates a 2 in 2 vertex line, and stores the information of all these lines in polar coordinates. It then searches for groups of lines whose number of occurrences (the number of lines of the same polar coordinates) is greater than a predetermined value, then use a "k-means" clustering algorithm (C122). , which will classify these groups of lines into 4 large groups (C123), each corresponding to one side of the rectangle formed by the correction template. Finally, we obtain an average value from each of the 4 groups, resulting in the lines that best represent each group, and intersecting these 4 lines (C124), we obtain the 4 vertices that give enough data to calculate a matrix. homographic (C12).

The software employing the method is applicable in the textile industry for digitizing molds, usually on paper. This computer program is therefore responsible for the digital reproduction of molds made and made by the modeler, and the reason for its existence is, above all, the preference of these professionals to perform their work in a manual / handcrafted manner, requiring later a program that interfaces these templates to CAD (Computer Aided Design) programs, where they will be enhanced, scaled, graduated, among other tasks.

The operator of the mold digitization process, illustrated in figure 4, has the function of fixing the molds to a digitization frame, which is surrounded by a correction template, indispensable for the orthogonalization and correction of optical distortions in the captured image. from the board. The captured image is then inserted into the digital computer that has the program installed, and in the latter, it is selected by the operator for mold extraction. From there, the correction method performs the entire sequence of procedures previously reported for image correction, without requiring interference or user input, finalizing the elaboration of an image without distortion.

With this "ideal" image, the program also automatically detects some features of the captured molds, such as their contours and inflection points. It also offers the operator some tools for other tasks, such as manual detection of some features that may or may not have been recognized by the software, contour corrections, etc.

At the end of the use of the software, a proprietary file is obtained, which is later used in a CAD software, seeking final hits as graduation of the molds in the most varied sizes.

Claims (4)

1. "AUTOMATED OPTICAL DISTORTION CORRECTION IMAGE FILE GENERATION PROCESS" characterized by: A) image capture - fixing objects inside the frame with a gated edge (A1) - capturing the image of objects and the template (A2) B) inserting the scanned image into the computer C) image processing and performing the distortion correction method of captured scanned images D) testing to verify that the image has been corrected E) final image correction - using tools to add and / or remove scanned features (E1) - CAD format file generation (E2) 2. "AUTOMATED OPTICAL DISTORTION CORRECTION IMAGE FILE GENERATION PROCESS" according to claim 1, comprising the method for correcting distortion of captured scanned images (C) by: data relating to the captured scanned image (B) in the software responsible for its execution; - performing the general procedure (C1) comprising: - locating the template vertices using the corner detection algorithm (C11); - using this data to make a first estimate of the homographic matrix (C12) which is a coordinate transformation matrix; - deletion of found corners that do not belong to the template (C13); - calculation of an improved estimate of homographic matrix (C14); - applying the homographic matrix to a copy of the detected vertices (C15), aiming to equate the corrected perspective vertices with data that would virtually be the exact location of the template vertices; - check (C16) whether a minimum number of points is successfully matched; - Optical distortion correction (C2) comprising: - Interaction in search of the coefficients of the equation variables which, when applied to the image, incurring a coordinate transformation to each of its pixels constituting a new image through the Least Squares algorithm (C21), such as the Levenberg-Marquardt nonlinear function minimizing algorithm; image correction for radial optical distortion (C22). orthogonalizing the image perspective (C3) comprising: - re-executing the general procedure (C1) from the first step (C11) to before the step removing the radial distortion (C2); - obtaining the orthogonalized image (C32). 3 "AUTOMATED OPTICAL DISTORTION CORRECTION IMAGE FILES PROCESS" according to claim 2, characterized in that the calculation of the homographic matrix (C12) comprises the following operations: - execution of the Hough Transform (C121), whose generated data define a line that best represents each of the 4 corners of the rectangle formed by the correction template, and for all vertices found by the corner detection algorithm (C11), the algorithm creates a 2 in 2 vertex line, and stores the information of all these lines in polar coordinates; - search for groups of lines whose number of occurrences (the number of lines of the same polar coordinates) is greater than a predetermined value, to then use a "k-means" clustering algorithm (C122); - classifying these straight groups into 4 large groups, each corresponding to one side of the rectangle (C123) formed by the correction template; - obtaining an average value from each of the 4 groups, with the lines that best represent each intersected group (C124) to obtain the 4 vertices that give sufficient data for the calculation of the homographic matrix (C12).