Coded disc surface tiny scratch identification method based on micro texture feature analysis

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a coded disc surface tiny scratch recognition method based on micro texture feature analysis, which comprises the following steps: acquiring a coded disc grating texture image and converting the coded disc grating texture image into a rectangular coordinate system to obtain an expanded grey-scale map; the optimal main direction of the grating texture is determined through rotation projection analysis; for each pixel point in the grey-scale map, sampling along two sides of the optimal main direction to form a reference sample; calculating a weighted average value of the gray values of the sample points in the reference sample to obtain a trend prediction value, calculating a residual error between the gray values of the pixel points and the trend prediction value, and determining a texture deviation index in combination with the change condition of the gray values of all neighborhood pixel points of the pixel points; and finally, identifying the scratches through threshold segmentation. According to the method, the grating edge interference can be effectively inhibited under the strong periodic texture background, and the tiny scratch features can be accurately extracted.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method for identifying minute scratches on the surface of a code disk based on micro-texture feature analysis. Background Technology

[0002] In the field of precision industrial control, the code disk of a photoelectric encoder is the core position feedback element. Its surface is etched with extremely high density radial grating textures. During the manufacturing process, due to the brittleness of the glass substrate and the complexity of the coating process, the disk surface is very prone to micron-level scratches, pinholes or dirt. These defects manifest as tiny damage to the periodic structure of the grating at the microscale, directly affecting the accuracy and lifespan of the encoder.

[0003] Currently, the industry mostly uses manual, carpet-like searches under microscopes, which is not only extremely inefficient, but also causes severe visual fatigue to the human eye due to prolonged observation of high-frequency stripes for faint scratches hidden in the dark lines of the grating or the transition area of ​​the edge, resulting in unavoidable missed detections and misjudgments.

[0004] Visual recognition algorithms, such as the Gabor filter algorithm, are often used to identify defects in textured surfaces such as textiles and wood because they can effectively extract textures at specific frequencies and directions. However, for special strongly periodic textured backgrounds like code disks, which are composed of dense grating stripes, the entire image is filled with high-frequency edge signals. From a signal processing perspective, normal grating edges and abnormal scratch edges have a serious overlap in the frequency domain. Traditional algorithms cannot effectively separate the two, which easily leads to a very high false alarm rate or serious missed detections, making it difficult to meet the needs of high-precision automated detection. Summary of the Invention

[0005] To address the technical problem of difficulty in detecting minute defects against a background of strong periodic texture on a code disk, and the easy confusion between grating edges and scratch edges, this invention provides a method for identifying minute scratches on the surface of a code disk based on micro-texture feature analysis. The method includes: acquiring a grating texture image of the code disk and transforming it from polar coordinates to Cartesian coordinates to obtain a grating unfolded grayscale image; rotating the grayscale image within a preset angle range with a set step size; acquiring the projection values ​​of all columns at each rotation angle; determining the probability that the rotation angle is the main direction of the texture based on the changes in the projection values ​​of all columns at any rotation angle; and setting the probability to the maximum value. A large rotation angle is determined as the optimal main direction of the raster texture. For each pixel in the grayscale image: multiple pixels are sampled along both sides of the optimal main direction to form a reference sample; the weighted average of the grayscale values ​​of each sample point in the reference sample is calculated to obtain the trend prediction value, and the weight of each sample point is inversely proportional to the distance from each sample point to the pixel; based on the residual between the grayscale value of the pixel and the trend prediction value, and the changes in the grayscale values ​​of all neighboring pixels of the pixel, the texture deviation index is determined; the texture deviation index of all pixels is thresholded, and pixels with a deviation index greater than the threshold are identified as scratch points.

[0006] This invention transforms complex radial grating textures into unfolded grayscale images using polar coordinate transformation. Based on the characteristics of projection value changes, it accurately locks the optimal principal direction of the texture. By utilizing the short-term stationarity of the grating along the principal direction, a trend prediction model is constructed. Combined with local neighborhood grayscale changes, the texture deviation index is calculated, achieving adaptive adjustment of detection sensitivity. This method effectively suppresses high-frequency grating background signals, significantly enhances the contrast between minute scratches and the background, and solves the problem that traditional algorithms struggle to distinguish grating edges from scratch edges and are prone to false alarms in strong periodic backgrounds, thus greatly improving recognition accuracy.

[0007] Preferably, the transformation from polar coordinates to rectangular coordinates to obtain the unfolded grayscale image includes: for a size of Any pixel in the unfolded grayscale image The coordinates of the object in the original image are calculated through reverse mapping. : , ,in This is the center position of the encoder disk. This represents the number of pixels from the center of the code disk to the innermost edge of the grating stripes. This represents the number of pixels along the circumferential tangent of the code disk. The radial width of the code disk is measured in pixels; this is obtained using an interpolation algorithm. The grayscale value at the location is assigned to .

[0008] This invention employs inverse mapping and interpolation algorithms for coordinate transformation, converting radial textures that vary with position into parallel textures with unified geometric features. This process eliminates the interference of differences in grating stripe angles, laying the foundation for constructing a unified one-dimensional prediction model and ensuring the effectiveness and consistency of subsequent predictions based on texture direction.

[0009] Preferably, obtaining the projection values ​​of all columns at each rotation angle includes: the projection value of each column at each rotation angle is equal to the sum of the gray values ​​of all pixels in each column of the expanded grayscale image after rotation at that rotation angle.

[0010] Preferably, the probability that the rotation angle is the main direction of the texture satisfies the expression: In the formula, For the first The probability that a rotation angle is the main direction of the texture; For the first At the nth rotation angle, the th The projected value of the column; For the first The mean of the projected values ​​of all columns at each rotation angle; To expand the grayscale image, specify the column indexes and the total number of columns in the horizontal direction.

[0011] This invention quantifies the probability of texture direction by calculating the variance of the projection values ​​of each column after rotation. It utilizes the principle that the separation of light and dark during stripe alignment maximizes the projection difference, which can accurately capture the angle of strongest signal coherence superposition, effectively correct the small deflection error introduced by mechanical or algorithmic means, and ensure that the prediction window extends strictly along the texture direction.

[0012] Preferably, the trend prediction value satisfies the expression: In the formula, For the first Trend prediction value for each pixel; For the first The nth pixel and its reference sample Euclidean distance between sample points; For the first In the reference sample of the nth pixel The grayscale value of each sample point; This refers to the index and total number of sample points in the reference sample sequence.

[0013] This invention establishes a trend prediction model based on inverse distance weighting. By assigning higher weights to sample points closer to the center pixel, it fully utilizes the strong spatial correlation of raster texture in the main direction. This bidirectional weighted statistics can accurately predict the theoretical value of the background, effectively smooth low-frequency changes caused by uneven illumination, and provide a precise benchmark for residual calculation.

[0014] Preferably, the texture deviation index satisfies the expression: In the formula, For the first Texture deviation index of each pixel; For the first Grayscale value and trend prediction value of each pixel; For the first The first pixel The grayscale value of each neighboring pixel; For the first The average grayscale value of all neighboring pixels of a given pixel; This represents the index and total number of pixels within the neighborhood; These are preset hyperparameters; This is the absolute value operator.

[0015] This invention normalizes the prediction residual by utilizing the local background noise level. In high-noise areas such as the grating edge, the larger background fluctuation rate suppresses the deviation exponent and avoids false alarms. In flat areas, even a small residual can produce a significant response. This adaptive mechanism achieves desensitization of high-frequency edge signals and enhancement of sensitivity to weak defect signals.

[0016] Preferably, the threshold segmentation is achieved using the maximum inter-class variance method.

[0017] Preferably, the preset angle range is: Set the step size as .

[0018] Preferably, all neighboring pixels of the pixel refer to all pixels within its eight neighboring pixels.

[0019] Preferably, the method further includes: performing morphological opening operations on the image formed by the identified scratch points to obtain the final scratch recognition result.

[0020] The beneficial effects of this invention are as follows: This invention predicts trends along the texture direction, canceling out normal variations in light and dark stripes as expected trends, thus completely eliminating the raster background in the result and retaining only the abrupt scratch signal. Even if the scratch has low contrast, as long as it causes a small jump in the statistical regularity of the pixel sequence, it will produce a non-negligible prediction residual, which will be captured by the algorithm. By introducing local volatility as a normalization factor, the detection sensitivity is automatically reduced at the raster edge and automatically increased in flat areas, effectively solving the problem of false alarms easily generated at the raster edge by traditional algorithms. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating a method for identifying minute scratches on the surface of a code disk based on micro-texture feature analysis, as described in this invention. Figure 2 This is a schematic representation of the grayscale image of the entire unfolded encoder disk grating; Figure 3 This is a schematic illustration of the scratch identification results after processing by the traditional Gabor filter algorithm; Figure 4 This is a schematic illustration of the scratch results identified after processing according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] This invention discloses a method for identifying minute scratches on the surface of a code disk based on micro-texture feature analysis, referring to... Figure 1 This includes steps S1 to S5: S1. Obtain the raster texture image of the code disk and transform it from polar coordinates to rectangular coordinates to obtain the raster unfolded grayscale image.

[0025] It should be noted that the physical stripes of the code disk are distributed radially around the center. If processed directly in the original Cartesian coordinate system, the angle of the grating stripes will change continuously with the position, making it impossible to build a unified one-dimensional prediction model. In order to transform the complex radial texture into a simple parallel texture, it is necessary to establish a mapping from the polar coordinate system to the rectangular coordinate system. This is the basis for subsequent predictions based on the short-term stationarity along the texture direction.

[0026] Specifically, a high-magnification industrial microscope camera is used to photograph the surface of the code disk to obtain a complete raster texture image, the position of each pixel in the image, and the center position of the code disk.

[0027] Furthermore, the size of the unfolded target image is defined as follows: ,in This represents the number of pixels along the circumferential tangent of the code disk. The number of pixels is the radial width of the encoder disk; for any point in the target image... The coordinates of the object in the original image are calculated through reverse mapping. : , ,in This is the center position of the encoder disk. The number of pixels from the center of the code disk to the innermost edge of the grating stripe; obtained using a bilinear interpolation algorithm. The grayscale value at that location is filled into the target image. Location.

[0028] It should be added that, as well as The value is determined based on the actual physical dimensions of the code disk and the microscope magnification, and the bilinear interpolation algorithm is a well-known technique, so it will not be elaborated here.

[0029] At this point, the unfolded grayscale image of the raster is obtained.

[0030] S2. Rotate the grayscale image within a preset angle range with a set step size; obtain the projection values ​​of all columns at each rotation angle; determine the probability that the rotation angle is the main direction of the texture based on the change of the projection values ​​of all columns at any rotation angle; determine the rotation angle with the highest probability as the optimal main direction of the raster texture.

[0031] It should be noted that while geometric mapping is used to convert the annular grating into a panoramic unfolded image, the unfolded grating texture is usually not able to present an ideal, strictly vertical distribution due to limitations such as the optical axis alignment deviation of the microscopic imaging system, the eccentric motion error of the mechanical rotating platform, and the discretization accuracy of the resampling algorithm. Instead, it appears as a tilted texture with a slight deflection angle. If prediction is made directly along the vertical direction, the prediction window will cross the light and dark boundaries of the grating, resulting in huge prediction errors in the normal area and thus causing serious misjudgments. Therefore, the actual direction of the texture is determined by finding the angle of strongest signal coherence superposition.

[0032] Specifically, preset angle range To set the step size Rotate the unfolded grayscale image; where, and The accuracy and efficiency of orientation locking are determined by this. Since the mechanical positioning error of precision equipment is usually very small, in the embodiments of this invention... Pick It can cover all working conditions. Pick , The smaller the value, the more accurate the orientation locking, the lower the residual baseline of the background prediction model established in subsequent steps, and the stronger the ability to detect subtle scratches. Implementers can balance this according to actual detection requirements.

[0033] For each rotation angle: obtain the sum of the gray values ​​of all pixels in each column of the expanded grayscale image after rotation as the projection value of each column; Based on the changes in the projection values ​​of all columns at any rotation angle, determine the probability that the rotation angle is the main direction of the texture; the probability that the rotation angle is the main direction of the texture satisfies the expression:

[0034] In the formula, For the first The probability that a rotation angle is the main direction of the texture; For the first At the nth rotation angle, the th The projected value of the column; For the first The mean of the projected values ​​of all columns at each rotation angle; To expand the grayscale image, specify the column indexes and the total number of columns in the horizontal direction.

[0035] in, Reflecting the At the nth rotation angle, the th The deviation of the projected value of the column from the overall mean, if the first column's projected value is... If the rotation angle is misaligned, the stripes will be tilted. During projection, bright and dark pixels will mix and cancel each other out in the same column, causing the cumulative values ​​of each column to tend to average. The value is relatively small; if the first When the rotation angle is precisely aligned, the stripes are perpendicular. At this point, the bright stripe columns are all high grayscale, and the dark stripe columns are all low grayscale, resulting in significant differences in the cumulative values ​​of each column. The value reaches its maximum.

[0036] Furthermore, iterate through all search angles and select the one that makes... The angle with the largest value is determined as the optimal main direction for the raster texture.

[0037] At this point, the optimal main direction of the raster texture has been obtained.

[0038] S3. For each pixel in the grayscale image: sample multiple pixels along both sides of the optimal main direction to form a reference sample; calculate the weighted average of the grayscale values ​​of each sample point in the reference sample to obtain the trend prediction value, and the weight of each sample point is inversely proportional to the distance of each sample point from the pixel.

[0039] It should be noted that in an ideal raster texture, the grayscale value of a pixel along the main direction of the texture should remain constant or show a slow, gradual change due to uneven illumination. This change belongs to low-frequency information. Scratches and other defects usually manifest as sudden disruptions to this continuity, which belong to high-frequency interference. Therefore, a two-way weighted statistical model can be constructed using the neighboring pixel samples in front of and behind the current pixel in the main direction of the texture to infer the theoretical background value of that position without defect interference.

[0040] Specifically, for any pixel in the raster unfolded grayscale image, along the optimal principal direction of the raster texture centered on that point, segments are taken from both sides. A number of pixels constitute a reference sample for that pixel; among them... The length of the reference sequence and the smoothing capability of the prediction model are determined in this embodiment of the invention. Pick The implementers can make adjustments based on the actual prediction quality, and for pixels where complete parameter samples cannot be obtained, the nearest neighbor interpolation method is used to fill in the gaps to ensure the integrity and continuity of the data.

[0041] Based on the distances of each sample point in the reference sample relative to this pixel, a weighted average of the gray values ​​of each sample point is calculated to determine the trend prediction value of this pixel; the trend prediction value satisfies the expression:

[0042] In the formula, For the first Trend prediction value for each pixel; For the first The nth pixel and its reference sample Euclidean distance between sample points; For the first In the reference sample of the nth pixel The grayscale value of each sample point; This refers to the index and total number of sample points in the reference sample sequence.

[0043] in, Reflecting the The reference sample sequence of the nth pixel The larger the weight coefficient of the i-th sample point, the more significant the weight coefficient. The closer each sample point is to the current center pixel, the stronger their spatial correlation in the direction of the raster texture extension, and the greater their indicative effect on the true gray value of the current pixel; 1 is added to the denominator to avoid the case where the distance is 0; this weight is used to calculate the weighted average of the gray values ​​in the reference samples, thereby obtaining the trend prediction value.

[0044] At this point, the trend prediction value of each pixel in the raster unfolded grayscale image has been obtained.

[0045] S4. Determine the texture deviation index based on the residual between the gray value of the pixel and the trend prediction value, as well as the changes in the gray values ​​of all neighboring pixels of the pixel.

[0046] It should be noted that using the difference between the original grayscale value and the trend prediction value as the scratch point is not rigorous enough. In the flat area of ​​the grating, i.e. the center of the bright or dark band, the signal is relatively stable, and a small residual may mean a defect. However, in the edge boundary area of ​​the grating, due to the point spread function of the camera and the sampling phase error, the pixel itself has drastic grayscale jumps and photon noise. In this area, a large residual may just be normal signal jitter. Therefore, the residual is normalized by using the noise level of the local background to achieve adaptive adjustment of the detection sensitivity.

[0047] Specifically, for any pixel in the unfolded grayscale image of the raster, all its eight neighboring pixels are obtained. Similarly, for pixels where the complete eight neighboring pixels cannot be obtained, the nearest neighbor interpolation method is used to fill the neighborhood.

[0048] Based on the residual between the grayscale value of the pixel and the trend prediction value, and the changes in the grayscale values ​​of all neighboring pixels of the pixel, the texture deviation index of the pixel is determined; the texture deviation index satisfies the expression:

[0049] In the formula, For the first Texture deviation index of each pixel; For the first Grayscale value and trend prediction value of each pixel; For the first The first pixel The grayscale value of each neighboring pixel; For the first The average grayscale value of all neighboring pixels of a given pixel; This represents the index and total number of pixels within the neighborhood; These are preset hyperparameters; This is the absolute value operator.

[0050] in, Reflecting the The grayscale value of a pixel is the residual between the prediction and the actual value. The larger the value, the more the grayscale value of the current pixel deviates significantly from the expected trend along the texture direction, which means that the continuity structure of the grating is broken at this point. Reflecting the The background volatility of a local region for each pixel indicates that the pixel is located at the edge of the light-dark boundary of the raster or in a high-noise region, indicating a high signal complexity of the background itself. In summary, if the original prediction residual is large and the local background volatility is small, then the calculated... A very large value indicates that the anomaly at that point has extremely high statistical significance; conversely, if it is in the marginal region, a large background volatility will act as a suppressor in the denominator. The value of is used to avoid false alarms at the edges; among which, To prevent the denominator from being zero, in the embodiments of the present invention, we take... .

[0051] At this point, the texture deviation index of each pixel in the raster unfolded grayscale image is obtained.

[0052] S5. Perform threshold segmentation on the texture deviation index of all pixels, and identify pixels with a deviation index greater than the segmentation threshold as scratch points.

[0053] It should be noted that the texture deviation index significantly highlights the scratch defect area. The deviation index of normal background texture is close to zero, while the deviation index of scratch area is significantly higher than that of background. Therefore, scratch area can be identified directly using existing threshold segmentation algorithms.

[0054] Specifically, the Otsu's method is used to calculate the segmentation threshold for the texture deviation index of all pixels in the raster unfolded grayscale image. Pixels with a texture deviation index greater than the segmentation threshold are identified as scratch defect points and their grayscale value is set to 255; pixels with a texture deviation index less than or equal to the segmentation threshold are identified as normal points and their grayscale value is set to 0.

[0055] Furthermore, morphological opening operations are performed on the image composed of the identified scratch points to obtain the final scratch recognition result; the morphological processing is to filter out any possible residual tiny isolated noise points and obtain clear, continuous scratches.

[0056] For example, Figure 2 The image is a grayscale image of the code disk grating unfolded in its entirety. In the periodic grating structure with alternating black and white areas, the black areas correspond to the dark bands of the grating, and the white areas correspond to the bright bands of the grating. Due to the uneven distribution of the light field in the microscopic illumination system, there is a certain gradation in the overall brightness of the image. In addition, there are tiny scratches in local areas of the grating. These defects appear to be subtle disruptions to the continuity of the vertical stripes under the naked eye, but these weak signals are easily ignored by the eye against the background of high-contrast black and white stripes. Figure 3 The scratch results identified by the traditional Gabor filter algorithm are filled with a large number of regularly distributed white vertical lines, which, while enhancing the scratches, also incorrectly enhance all the grating edges, resulting in serious false alarms in the detection results. Figure 4 The image shows the scratch identification results after processing by this invention. The background is pure black, indicating that normal raster stripe backgrounds have been suppressed by the algorithm. Simultaneously, white connected regions are clearly visible in the image, and these white areas correspond to... Figure 2 Compared to traditional frequency domain filtering methods, this method has stronger background suppression and weak defect detection capabilities for even the smallest scratch locations, significantly improving the accuracy and robustness of detection.

Claims

1. A code disc surface micro-scratch identification method based on micro-texture feature analysis, characterized in that, The method comprises the following steps: an optical grating texture image of a code disc is acquired, and a polar coordinate to rectangular coordinate system transformation is performed to obtain an optical grating expansion grayscale image; the grayscale image is rotated at a set step within a preset angle range; projection values of all columns at each rotation angle are acquired; a probability that the rotation angle is a main direction of the texture is determined according to a change condition of the projection values of all columns at the rotation angle; and the rotation angle with the maximum probability is determined as the optimal main direction of the optical grating texture; for each pixel point in the grayscale image: a plurality of pixel points on both sides of the optimal main direction are sampled to form a reference sample; a weighted average value of grayscale values of sample points in the reference sample is calculated to obtain a trend prediction value, and the weight of each sample point is inversely proportional to the distance between the sample point and the pixel point; a texture deviation index is determined according to a residual error between the grayscale value and the trend prediction value of the pixel point and a change condition of grayscale values of all neighborhood pixel points of the pixel point; and threshold segmentation is performed on the texture deviation indexes of all pixel points, and a pixel point greater than a segmentation threshold is identified as a scratch point.

2. The method according to claim 1, wherein, The polar coordinate to rectangular coordinate system transformation to obtain the optical grating expansion grayscale image comprises the following steps: For any pixel point in the expanded gray image with size , its coordinate in the original image is calculated by inverse mapping : , wherein is the center position of the code disc, is the pixel number from the center of the code disc to the innermost edge of the grating stripe; is the pixel number of the circumferential tangent of the code disc, is the pixel number of the radial width of the code disc; the gray value at is obtained by using an interpolation algorithm and is assigned to .​ 3. The method according to claim 1, wherein, The acquisition of the projection values of all columns at each rotation angle comprises the following steps: the projection value of each column at each rotation angle is equal to the sum of grayscale values of all pixel points of each column in the expansion grayscale image after rotation at the rotation angle.

4. The method according to claim 1, wherein, The probability that the rotation angle is the main direction of the texture satisfies the expression: ； In the formula, is the probability of the texture main direction of the first rotation angle; is the projection value of the first column under the first rotation angle; is the mean value of the projection values of all columns under the first rotation angle; is the column index and the total number of columns in the expanded grayscale image in the horizontal direction.

5. The method according to claim 1, wherein, The trend prediction value satisfies the expression: ； In the formula, is the trend prediction value of the first pixel point; is the trend prediction value of the first pixel point; is the Euclidean distance between the first pixel point and the first sample point in the reference sample of the first pixel point; is the gray value of the first sample point in the reference sample of the first pixel point; and is the index and total number of the sample points in the reference sample sequence.

6. The method according to claim 1, wherein, The texture deviation index satisfies the expression: ; In the formula, For the first Texture deviation index of each pixel; For the first Grayscale value and trend prediction value of each pixel; For the first The first pixel The grayscale value of each neighboring pixel; For the first The average grayscale value of all neighboring pixels of a given pixel; This represents the index and total number of pixels within the neighborhood; These are preset hyperparameters; This is the absolute value operator.

7. The method according to claim 1, wherein, The threshold segmentation is realized by a maximum inter-class variance method.

8. The method according to claim 1, wherein, The preset angle range is , the step length is set as .

9. The method according to claim 1, wherein, All neighborhood pixel points of the pixel point refer to all pixel points in the eight neighborhoods of the pixel point.

10. The method according to claim 1, wherein, The method further comprises the following steps: a morphological opening operation processing is performed on an image composed of the identified scratch points to obtain a final scratch recognition result.

Images