Data line appearance defect rapid detection method based on intelligent image recognition

Abstract

The invention discloses a data line appearance defect rapid detection method based on intelligent image recognition, relates to the technical field of image processing, and aims to solve the technical problem that part of defects are omitted due to limited image acquisition angles in the prior art. Comprising the following steps: S1, setting a plurality of groups of image acquisition modules which are distributed in an annular array by taking a data line as a center, and setting illumination modules on two sides and the middle part of each group of image acquisition modules; s2, performing multi-angle image acquisition on the data line passing through the detection area at a constant speed by using an image acquisition module, and shooting and recording the position of the data line according to a preset time interval; s3, denoising, contrast enhancement and graying processing are carried out on the acquired image; s4, feature points of the preprocessed images are extracted and matched, and after a transformation matrix is calculated, the images are spliced into a data line surface overall image; and S5, comparing the overall image with a qualified data line standard image. The method has the advantage of improving the completeness and accuracy of defect detection.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and more specifically to a method for quickly detecting appearance defects of a data line based on intelligent image recognition. Background Art

[0002] In the data cable production process, data cable appearance defect detection is an important link to ensure product quality. The traditional data cable appearance defect detection method mainly relies on manual inspection, where inspectors observe with the naked eye whether there are defects such as convex hulls, concave dents, scratches, etc. on the surface of the data cable. This manual inspection method is not only inefficient and difficult to meet the inspection needs of large-scale production, but the inspection results are easily affected by the subjective factors of the inspectors, such as fatigue and emotions, resulting in low inspection accuracy, and missed inspections and false inspections occur from time to time.

[0003] With the development of computer technology and image processing technology, detection methods based on image recognition have gradually been applied to the field of data line appearance defect detection. Existing detection methods based on image recognition usually set up a single image acquisition device at a fixed position, shoot the data line, and then judge whether there are defects through simple image comparison or feature extraction. Although this method improves the detection efficiency to a certain extent, it is difficult to fully obtain the information on the surface of the data line due to the limited angle of the image collected, and it is easy to miss some defects at special angles.

[0004] In view of this, we propose a rapid detection method for data line appearance defects based on intelligent image recognition. Summary of the invention

[0005] The purpose of the present invention is to provide a method for quickly detecting appearance defects of data lines based on intelligent image recognition, so as to solve the technical problem in the prior art that some defects are missed due to the limited image acquisition angle.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for rapid detection of data line appearance defects based on intelligent image recognition, comprising the following steps: S1. Arrange a plurality of groups of image acquisition modules distributed in a circular array with the data line as the center, and arrange illumination modules on both sides and in the middle of each group of image acquisition modules; S2, using the image acquisition module to collect multi-angle images of the data lines passing through the detection area at a uniform speed, and taking and recording the positions of the data lines at preset time intervals; S3, denoising, contrast enhancement and grayscale processing of the collected images; S4, extracting and matching the feature points of the preprocessed image, and splicing the image into an overall image of the data line surface after calculating the transformation matrix; S5, comparing the overall image with the qualified data line standard image, and calculating the gray value difference of the corresponding pixel points; S6. Determine the convex hull or concave area and its position on the surface of the data line according to the gray value difference; S7, start the corresponding illumination modules to fill in the defect position in sequence, and collect the image of the defect position again; S8, analyzing the gray value change of the image after the fill light, and calculating the convex hull height and the concave depth respectively; S9. Compare the calculated convex hull height and concave depth with their corresponding thresholds respectively, and classify the data lines into qualified products, defective products or waste products.

[0007] Preferably, in S3, a Gaussian filtering algorithm is used for denoising, assuming that the original image is , the denoised image Calculated by the following formula: ; in, is the Gaussian kernel standard deviation, which is adaptively adjusted according to the noise level and resolution of the image to achieve the best denoising effect. and is the integral variable, which is used to perform weighted summation of the original image at different positions; Adaptive histogram equalization algorithm is used to enhance the image contrast. Assume the image gray level is , the original image pixel value is , the probability of occurrence is Pixel values ​​after adaptive histogram equalization Calculated by the following formula: ; in, is the total number of pixels in the image, is the local area currently being processed, is the Dirac function.

[0008] Preferably, in S4, the image stitching method is: The SIFT algorithm is used to extract the feature points of each group of images. The scale space is , through the Gaussian difference function To detect extreme points, the calculation formula is: ; in, is a Gaussian function, is the scale factor; The FLANN algorithm is used to match the feature points of different groups of images, and a fast search is performed based on the KD tree data structure. The feature point set is , when searching for the nearest neighbor, the Euclidean distance is calculated To determine, the calculation formula is: ; in, is the feature point vector dimension, and Represents the feature points and In the The coordinate value of the dimension; Use the RANSAC algorithm to calculate the transformation matrix between images, assuming that For the sample, calculate the homography matrix , so that satisfaction has the largest number of matching point pairs, among which, is the feature point of the original image, is the corresponding feature point of the target image.

[0009] Preferably, different groups of images are spliced ​​according to the transformation matrix, and the splicing seams are eliminated by a fade-in and fade-out fusion method. After the splicing is completed, the quality of the spliced ​​image is evaluated by calculating the clarity of the spliced ​​area. If the quality does not meet the preset standard, feature point extraction, matching and splicing operations are performed again. The clarity and contrast are evaluated by calculating the gradient amplitude, respectively. and the local standard deviation The calculation formula is: ; ; in, is the local area mean.

[0010] Preferably, in S5, a template matching algorithm is first used to roughly match the whole image with the standard image. Assume that the template image is , the search image is , by calculating the normalized cross-correlation coefficient Determine the approximate matching area, the calculation formula is: ; in, For the coordinates Template Image The normalized correlation coefficient with the search image is used to measure the degree of match between the two. The template image is at coordinates The pixel value at is the mean of the template image, To search for an image at coordinates The pixel value at is the mean of the search image area corresponding to the template image, and is the size of the template image, that is, the template image is and The number of pixels in the direction; Then, the gray value difference of the corresponding pixel points is calculated by pixel-by-pixel subtraction in the area. , and calculate the mean of the gray value differences and variance Statistical features, among which, and To match the area image in and The number of pixels in the direction.

[0011] Preferably, in S6, a gray value difference threshold is set based on a large amount of experimental data and actual production requirements. , when the gray value difference of a pixel When the threshold is exceeded, it is marked as a suspected defect point; Perform eight-neighborhood connectivity analysis on the marked pixels, and determine the area consisting of interconnected marked pixels as the defect area; Calculate the area, perimeter, and center of gravity of each defect area to further clarify the location and shape of the defect; Morphological processing is performed on the identified defective areas using dilation and erosion operations; Remove noise and small interference areas, and optimize the shape and boundaries of defect areas; Calculate the second-order central moment of the grayscale value of the pixel in the defect area , , , the calculation formula is: ; in, , , is the second-order central moment of the grayscale value of the pixel in the defect area, which is used to describe the characteristics of the grayscale value distribution in the defect area; Calculate the directional characteristics of the defect area , and its calculation formula is: ; Center of gravity Centered along the direction and its vertical direction Take two straight lines and , the defect area is divided into four sub-areas , , , ; Calculate the sum of the gray value differences in the four sub-areas , , , ; like , then the defect area is judged as a convex hull, otherwise, it is judged as a concave.

[0012] Preferably, in S7, the average grayscale values ​​in the convex hull and the concave area under different fill light angles are counted respectively, and the influence of abnormal values ​​is removed by using median filtering. The pixel point set in the convex hull area after the second fill light is: , average gray value , the pixel point set in the concave area is , average gray value ; Compare the changes in the average grayscale values ​​of the convex hull and concave areas under different fill light angles, draw a curve of the grayscale value changing with the fill light angle, and calculate the slope of the curve to obtain the grayscale gradient change; Assume that the two adjacent fill light angles are , The convex hull corresponds to the average gray value of , , grayscale gradient ; The average gray value corresponding to the concave is , , grayscale gradient ; Assume that the convex hull height mapping function is , the depression depth mapping function is ; The linear interpolation analysis method is used to calculate the height of the convex hull and the depth of the concave according to the gray gradient change.

[0013] Preferably, for the convex hull, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated convex hull height is for: ; For the concavity, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated depression depth for: .

[0014] Preferably, in S9, the data line classification rule is: Let the convex hull height threshold be , the convex hull rejection threshold is , the depression depth threshold is , the concave rejection threshold is ; like and , then the data cable is judged to be a qualified product; like , then the data cable is judged to be defective; like , then the data cable is judged to be scrap.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention realizes the effect of all-round acquisition of data line images by designing an image acquisition module structure distributed in a ring array. Compared with the traditional single image acquisition device, the present invention can obtain more comprehensive data line surface information, greatly improve the integrity of defect detection, and solve the problem of missing some defects due to limited image acquisition angle in the prior art.

[0016] 2. The present invention also realizes rapid detection by photographing and recording the positions of data lines passing through the detection area at a uniform speed at preset time intervals, and cooperating with efficient image processing algorithms. In the data processing process, SIFT, FLANN and other algorithms are used to quickly extract and match feature points, and image stitching and defect judgment operations can be completed in a short time, further solving the problem of low efficiency of manual detection, meeting the demand for rapid detection of data lines in large-scale production, and improving production efficiency.

[0017] 3. The present invention utilizes SIFT algorithm, FLANN algorithm and RANSAC algorithm to extract feature points, match and stitch images, and calculates convex hull height and concave depth by analyzing the change of image grayscale value after fill light, and classifies data lines by comparing with threshold values. This further solves the problem of inaccurate calculation of defect depth and height, inability to accurately evaluate and classify, and can make more accurate judgments on data line quality, meeting the needs of manufacturers for strict control of product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0019] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0020] Example 1

[0021] like Figure 1 As shown, the present invention provides a method for rapid detection of data line appearance defects based on intelligent image recognition, which is characterized by comprising the following steps: S1. Arrange a plurality of groups of image acquisition modules distributed in a circular array with the data line as the center, and arrange illumination modules on both sides and in the middle of each group of image acquisition modules; S2, using the image acquisition module to collect multi-angle images of the data lines passing through the detection area at a uniform speed, and taking and recording the positions of the data lines at preset time intervals; S3, denoising, contrast enhancement and grayscale processing of the collected images; S4, extracting and matching the feature points of the preprocessed image, and splicing the image into an overall image of the data line surface after calculating the transformation matrix; S5, comparing the overall image with the qualified data line standard image, and calculating the gray value difference of the corresponding pixel points; S6. Determine the convex hull or concave area and its position on the surface of the data line according to the gray value difference; S7, start the corresponding illumination modules to fill in the defect position in sequence, and collect the image of the defect position again; S8, analyzing the gray value change of the image after the fill light, and calculating the convex hull height and the concave depth respectively; S9. Compare the calculated convex hull height and concave depth with their corresponding thresholds respectively, and classify the data lines into qualified products, defective products or waste products.

[0022] In an embodiment of the present invention, in S1, the angles between adjacent image acquisition modules are equal and evenly distributed in the circumferential direction of the data line, ensuring that the data line image can be acquired in all directions, and the illumination module is an LED light source with adjustable brightness, which can dynamically adjust the brightness and angle of the illumination according to the material and surface characteristics of the data line, thereby providing the best illumination conditions for image acquisition.

[0023] In the embodiment of the present invention, in S2, the data line is at a constant speed Through the detection area, the image acquisition module follows the preset fixed time interval Take photos and monitor and record the position of the data line in real time. .

[0024] In the embodiment of the present invention, in S3, Gaussian filtering algorithm is used for denoising. Assume that the original image is , the denoised image Calculated by the following formula: ; in, is the Gaussian kernel standard deviation, which is adaptively adjusted according to the noise level and resolution of the image to achieve the best denoising effect. and is the integral variable, which is used to perform weighted summation of the original image at different positions; Adaptive histogram equalization algorithm is used to enhance the image contrast. Assume the image gray level is , the original image pixel value is , the probability of occurrence is Pixel values ​​after adaptive histogram equalization Calculated by the following formula: ; in, is the total number of pixels in the image, is the local area currently being processed, is the Dirac function.

[0025] In an embodiment of the present invention, in S4, the image stitching method is: The SIFT algorithm is used to extract the feature points of each group of images. The scale space is , through the Gaussian difference function To detect extreme points, the calculation formula is: ; in, is a Gaussian function, is the scale factor; The FLANN algorithm is used to match the feature points of different groups of images, and a fast search is performed based on the KD tree data structure. The feature point set is , when searching for the nearest neighbor, the Euclidean distance is calculated To determine, the calculation formula is: ; in, is the feature point vector dimension, and Represents the feature points and In the The coordinate value of the dimension; Use the RANSAC algorithm to calculate the transformation matrix between images, assuming that For the sample, calculate the homography matrix , so that satisfaction has the largest number of matching point pairs, among which, is the feature point of the original image, Corresponding feature points of the target image; Different groups of images are stitched together according to the transformation matrix, and the stitching seams are eliminated by the fade-in and fade-out fusion method. After the stitching is completed, the quality of the stitched image is evaluated by calculating the clarity of the stitching area. If the quality does not meet the preset standard, the feature point extraction, matching and stitching operations are performed again. The clarity and contrast are evaluated by calculating the gradient amplitude respectively. and the local standard deviation The calculation formula is: ; ; in, is the local area mean.

[0026] In the embodiment of the present invention, in S5, a template matching algorithm is first used to roughly match the whole image and the standard image. Assume that the template image is , the search image is , by calculating the normalized cross-correlation coefficient Determine the approximate matching area, the calculation formula is: ; in, For the coordinates Template Image The normalized correlation coefficient with the search image is used to measure the degree of match between the two. The template image is at coordinates The pixel value at is the mean of the template image, To search for an image at coordinates The pixel value at is the mean of the search image area corresponding to the template image, and is the size of the template image, that is, the template image is and The number of pixels in the direction; Then, the gray value difference of the corresponding pixel points is calculated by pixel-by-pixel subtraction in the area. , and calculate the mean of the gray value differences and variance Statistical features, among which, and To match the area image in and The number of pixels in the direction.

[0027] In the embodiment of the present invention, in S6, a gray value difference threshold value is set based on a large amount of experimental data and actual production requirements. , when the gray value difference of a pixel When the threshold is exceeded, it is marked as a suspected defect point; Perform eight-neighborhood connectivity analysis on the marked pixel points, assuming that the current pixel point is , and its eight neighboring pixels are , and ,like , then the neighborhood point is connected to the current point, and the area composed of the interconnected marked pixel points is determined as the defect area; Calculate the area of ​​each defect region ,perimeter , Center of gravity , further clarify the location and shape of the defects; The determined defect area is morphologically processed by using the dilation operation, and the structural element is set as , the expanded image for: ; During the erosion operation, the eroded image for: ; Remove noise and small interference areas, and optimize the shape and boundaries of defect areas; Calculate the second-order central moment of the grayscale value of the pixel in the defect area , , , the calculation formula is: ; in, , , is the second-order central moment of the grayscale value of the pixel in the defect area, which is used to describe the characteristics of the grayscale value distribution in the defect area; Calculate the directional characteristics of the defect area , and its calculation formula is: ; Center of gravity Centered along the direction and its vertical direction Take two straight lines and , the defect area is divided into four sub-areas , , , ; Calculate the sum of the gray value differences in the four sub-areas , , , ; like , then the defect area is judged as a convex hull, otherwise, it is judged as a concave.

[0028] In the embodiment of the present invention, in S7, the average grayscale values ​​in the convex hull and the concave area under different fill light angles are counted respectively, and the influence of abnormal values ​​is removed by median filtering. The pixel point set in the convex hull area after the second fill light is: , average gray value , the pixel point set in the concave area is , average gray value ; Compare the changes in the average grayscale values ​​of the convex hull and concave areas under different fill light angles, draw a curve of the grayscale value changing with the fill light angle, and calculate the slope of the curve to obtain the grayscale gradient change; Assume that the two adjacent fill light angles are , The convex hull corresponds to the average gray value of , , grayscale gradient ; The average gray value corresponding to the concave is , , grayscale gradient ; Assume that the convex hull height mapping function is , the depression depth mapping function is ; The linear interpolation analysis method is used to calculate the height of the convex hull and the depth of the concave according to the gray gradient change; For the convex hull, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated convex hull height is for: ; For the concavity, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated depression depth for: ; Considering the attenuation of light intensity and the influence of the surface reflectivity of the data line, a light attenuation model is established. The distance between the light source and the defect position is , the initial light intensity is , the attenuated light intensity , is the attenuation coefficient; establish a reflectivity model, assuming that the surface reflectivity of the data line is , the actual received light intensity , correct the gray value; The gray value of the corrected convex hull area is , ; The gray value of the corrected concave area is , , improving the accuracy of height and depth calculations.

[0029] In the embodiment of the present invention, in S9, the data line classification rule is: Let the convex hull height threshold be , the convex hull rejection threshold is , the depression depth threshold is , the concave rejection threshold is ; like and , then the data cable is judged to be a qualified product; like , then the data cable is judged to be defective; like , then the data cable is judged to be scrap.

[0030] The embodiments of the present invention disclose preferred embodiments, but are not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A method for rapid detection of data line appearance defects based on intelligent image recognition, characterized in that: The following steps are involved: S1. Arrange a plurality of groups of image acquisition modules distributed in a circular array with the data line as the center, and arrange illumination modules on both sides and in the middle of each group of image acquisition modules; S2, using the image acquisition module to collect multi-angle images of the data lines passing through the detection area at a uniform speed, and taking and recording the positions of the data lines at preset time intervals; S3, denoising, contrast enhancement and grayscale processing of the collected images; S4, extracting and matching the feature points of the preprocessed image, and splicing the image into an overall image of the data line surface after calculating the transformation matrix; S5, comparing the overall image with the qualified data line standard image, and calculating the gray value difference of the corresponding pixel points; S6. Determine the convex hull or concave area and its position on the surface of the data line according to the gray value difference; S7, start the corresponding illumination modules to fill in the defect position in sequence, and collect the image of the defect position again; S8, analyzing the gray value change of the image after the fill light, and calculating the convex hull height and the concave depth respectively; S9. Compare the calculated convex hull height and concave depth with their corresponding thresholds respectively, and classify the data lines into qualified products, defective products or waste products.

2. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 1, characterized in that: In S3, Gaussian filtering algorithm is used for denoising. Suppose the original image is , the denoised image Calculated by the following formula: ； in, is the Gaussian kernel standard deviation, which is adaptively adjusted according to the noise level and resolution of the image to achieve the best denoising effect. and is the integral variable, which is used to perform weighted summation of the original image at different positions; Adaptive histogram equalization algorithm is used to enhance the image contrast. Assume the image gray level is , the original image pixel value is , the probability of occurrence is Pixel values ​​after adaptive histogram equalization Calculated by the following formula: ； in, is the total number of pixels in the image, is the local area currently being processed, is the Dirac function.

3. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 2, characterized in that: In S4, the image stitching method is: The SIFT algorithm is used to extract the feature points of each group of images. The scale space is , through the Gaussian difference function To detect extreme points, the calculation formula is: ； in, is a Gaussian function, is the scale factor; The FLANN algorithm is used to match the feature points of different groups of images, and a fast search is performed based on the KD tree data structure. The feature point set is , when searching for the nearest neighbor, the Euclidean distance is calculated To determine, the calculation formula is: ； in, is the feature point vector dimension, and Represents the feature points and In the The coordinate value of the dimension; Use the RANSAC algorithm to calculate the transformation matrix between images, assuming that For the sample, calculate the homography matrix , so that satisfaction has the largest number of matching point pairs, among which, is the feature point of the original image, is the corresponding feature point of the target image.

4. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 3, characterized in that: Different groups of images are stitched together according to the transformation matrix, and the stitching seams are eliminated by the fade-in and fade-out fusion method. After the stitching is completed, the quality of the stitched image is evaluated by calculating the clarity of the stitching area. If the quality does not meet the preset standard, the feature point extraction, matching and stitching operations are performed again. The clarity and contrast are evaluated by calculating the gradient amplitude respectively. and the local standard deviation The calculation formula is: ； ； in, is the local area mean.

5. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 4, characterized in that: In S5, a template matching algorithm is first used to roughly match the whole image with the standard image. Suppose the template image is , the search image is , by calculating the normalized cross-correlation coefficient Determine the approximate matching area, the calculation formula is: ； in, For the coordinates Template Image The normalized correlation coefficient with the search image is used to measure the degree of match between the two. The template image is at coordinates The pixel value at is the mean of the template image, To search for an image at coordinates The pixel value at is the mean of the search image area corresponding to the template image, and is the size of the template image, that is, the template image is and The number of pixels in the direction; Then, the gray value difference of the corresponding pixel points is calculated by pixel-by-pixel subtraction in the area. , and calculate the mean of the gray value differences and variance Statistical features, among which, and To match the area image in and The number of pixels in the direction.

6. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 5, characterized in that: In S6, a gray value difference threshold is set based on a large amount of experimental data and actual production requirements. , when the gray value difference of a pixel When the threshold is exceeded, it is marked as a suspected defect point; Perform eight-neighborhood connectivity analysis on the marked pixels, and determine the area consisting of interconnected marked pixels as the defect area; Calculate the area, perimeter, and center of gravity of each defect area to further clarify the location and shape of the defect; Morphological processing is performed on the identified defective areas using dilation and erosion operations; Remove noise and small interference areas, and optimize the shape and boundaries of defect areas; Calculate the second-order central moment of the grayscale value of the pixel in the defect area , , , the calculation formula is: ； in, , , is the second-order central moment of the grayscale value of the pixel in the defect area, which is used to describe the characteristics of the grayscale value distribution in the defect area; Calculate the directional characteristics of the defect area , and its calculation formula is: ； Center of gravity Centered along the direction and its vertical direction Take two straight lines and , the defect area is divided into four sub-areas , , , ; Calculate the sum of the gray value differences in the four sub-areas , , , ; like , then the defect area is judged as a convex hull, otherwise, it is judged as a concave.

7. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 6, characterized in that: In S7, the average grayscale values ​​in the convex hull and concave regions under different fill light angles are counted respectively, and the influence of abnormal values ​​is removed by median filtering. The pixel point set in the convex hull area after the second fill light is: , average gray value , the pixel point set in the concave area is , average gray value ; Compare the changes in the average grayscale values ​​of the convex hull and concave areas under different fill light angles, draw a curve of the grayscale value changing with the fill light angle, and calculate the slope of the curve to obtain the grayscale gradient change; Assume that the two adjacent fill light angles are , The convex hull corresponds to the average gray value of , , grayscale gradient ; The average gray value corresponding to the concave is , , grayscale gradient ; Assume that the convex hull height mapping function is , the depression depth mapping function is ; The linear interpolation analysis method is used to calculate the height of the convex hull and the depth of the concave according to the gray gradient change.

8. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 7, characterized in that: For the convex hull, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated convex hull height is for: ； For the concavity, the grayscale gradient is known Falling in the range , the corresponding height value is , , then the calculated depression depth for: 。 9. The method for rapid detection of data line appearance defects based on intelligent image recognition according to claim 8, characterized in that: In S9, the data line classification rules are: Let the convex hull height threshold be , the convex hull rejection threshold is , the depression depth threshold is , the concave rejection threshold is ; like and , then the data cable is judged to be a qualified product; like , then the data cable is judged to be defective; like , then the data cable is judged to be scrap.

Images