Research On Leather Scratch Detection Method Based On Texture Analysis

With the continuous improvement of people’s life,leather products are increasingly getting access to people’s lives.However,the scratches of leather,which reduce the service life of leather products,can not only affect the aesthetics of it,but also cause safety hazards.Therefore,it’s significant to detect the scratches on the surface of leather products.In recent years,machine vision technology has been widely used in various fields,making it a popular choice to replace traditional manual quality inspection with machines.Thus this thesis mainly applies computer image processing technology to the detection of scratches on the surface of lychee leather.Aiming at the problem that the traditional Gauss-Laplace algorithm has the isotropic characteristics when processing image edges,an anisotropic Gauss-Laplace algorithm is proposed to adaptively process the edges of various angles in image enhancement.Firstly,on the base of traditional operator,different Gaussian standard deviations are taken in relevant directions to introduce the scale parameters into the function.Then,the angle parameters are introduced into the operator with scale parameters in combination with geometry.And finally,an anisotropic operator is obtained through determining the values of the scale and angle parameters according to the gradient value of each pixel.The simulation results show that the anisotropic operator is more suitable for human visual characteristics,and is superior to isotropic operators and other image preprocessing algorithms in terms of peak signal-to-noise ratio,structural similarity and mean square error.The anisotropic Gauss-Laplace has better noise suppression capability and remains more edge details of the image.To solve the problem that the current single texture analysis method is unable to accurately extract the image texture primitive information during image processing,a variety of texture analysis methods,each one plays its own characteristics in texture analysis,are combined to maximize the extraction of texture primitive information.Thereby the accuracy of the detection results is improved.Firstly,wavelet decomposition is used to obtain the low-low frequencies of the image,which contain more texture information.Through experimental comparison and analysis,the optimal characteristic parameters for the detection of striated leather scratches were selected to determine the three optimal construction factors of the gray level co-occurrence matrix.Moreover,the corresponding gray level co-occurrence matrix is generated.Finally,corresponding features are extracted from the matrix containing optimal parameters,and the surface scratches of the leather are detected by using a support vector machine.According to the surface texture characteristics of lychee leather,the corresponding machine vision inspection platform is built.Comparison experiments indicate that the detection accuracy of multi-method combination which simultaneously takes real-time performance into account is 11.93% higher than wavelet transform method,1.33% higher than gray level co-occurrence matrix method.And the proposed algorithm applying a variety of texture analysis methods is proved to be effect in the detection of surface scratches on pebbled leather,the final detection accuracy is 95.2%,and the single image detection time is 372 ms.