Evaluating The Rate Of Fabric's Waterproof Performance Based On Digital Image Processing

The evaluation of fabric's waterproof performance has always been an important content in the field of textile testing,since the waterproof performance not only is an important indicator of the quality of the waterproof fabric but also reduce the wearing comfort.Traditionally,the rate of fabric's waterproof performance is judged according to the standard sample.But it is greatly subjective and will produce different results among different people.To enhance the objectivity of fabric's waterproof performance evaluation,using digital image processing technique to evaluate the rate of fabric's waterproof performance automatically is of great significance.The main contents are as follows:?1?The contrast preserving decolorization algorithm is chosen to process wetting image in advance,so as to preserve original color contrast between wetting region and background,preventing from over segmentation during extract wetting region.Simultaneously,apply adaptive histogram equalization to avoid excessive noise or shadow in the background area..Current automatic detection of fabric waterproof performance can not eliminate the noise interference and suppress the influence of uneven illumination.Thus,this paper proposes the combination of L₀ gradient minimization and laplacian eigenmaps for image smoothing to remove noise interference.According to the edge information of fabric wetting image,the gradient of each pixel in a image is calculated along the x and y directions.With the number of non-zero gradients constraint,the fabric wetting image can be smoothed while sharpening major edges.Then map the image into CIELab color space to adjust the L component to eliminate the effect of uneven illumination.In a neighborhood,calculate the affinity between pixels to array into affinity matrix.Meanwhile.Based on laplacian eigenmaps theory,a piecewise image consisting of calculated eigenvectors is converted to a smoothing approximation image by fast Fourier transform.Finally,the segmentation image is obtained by fuzzy clustering.The experimental results show that this method can effectively achieve denoising treatment of extracting wetting region of fabric,and the segmentation image of fabric wetting region isaccurately gained.?2?As the shapes of the extracted wetting regions are different,it is necessary to convert the extracted image to some form adapted to the image classification task.Exploit SIFT algorithm to extract descriptors from a segmentation image exclusively containing wetting region.According to sparse representation,the coding coefficients are obtained from the codebook consisting of descriptors.Based on the above two steps,the feature vector is a concatenation of coefficients by multi-scale spatial max pooling coding coefficients.The smoothing algorithm proposed in this paper can not be completely removed for the stubborn spot reflection noise of sparse distribution,that is,the feature vector extracted from the segmentation image contains noise.Because the segmentation images are low-rank and the eigenvectors can be represented sparsely by the dictionary,in order to make the atoms in the discriminative dictionary more pure,the coding coefficients that can represent the eigenvectors are constrained by low rank.When the discriminative dictionary is updated,the label consistent constraint is added to the sub-dictionary and the coding coefficient.With the current coding coefficients of dictionary,we address this problem by using the augmented Lagrange multiplier method.With the current dictionary,we exploit SVD decomposition to update dictionary.Finally,an discriminative dictionary is obtained.Experiments show that the discriminative dictionary with low rank constraint can get 96%accuracy when evaluate the fabric's waterproof performance.