Research On Fabric Image Defect Detection Algorithm Based On S Transform

Improving production quality and lowering production costs are the key points for the textile enterprise survival and the development. And fabric defect detection is the key to quality control. Fabric surface shows regular texture characteristics and there is no rigid definitions method for fabric defect. Therefore fabric defect detection is a challenging task.Fabric texture, which have a strong periodicity in different directions, is formed by the repetition of a basic texture element. It has a strong periodicity in all directions, and the value on the spectral response curve at the corresponding frequency point is large. The presence of defects destroy the fabric texture periodic, lead to the spectrum changes at the corresponding points. Therefore, some defect detection methods based on spectral analysis have been proposed in many documents. It is shown that the multi-resolution analysis methods such as Gabor filters, Fourier transform and wavelet transform are beneficial for fabric defect detection.S transform is a relatively new non-stationary time-frequency signal analysis method, which is a combination of short-time Fourier transform and wavelet transform. It can clearly characterize the instantaneous frequency features, and has some advantages such as strong anti-noise, no cross-term interference, simple calculation and intuitive expression. In this paper, S transform, which is a new relatively signal analysis method, is adopted for describing the normal texture features and the defects features, and is applied for defect detection. The main contents can be listed as:(1) Adopted a new method for fabric defect identification based on the gray projection and one-dimensional S transform. Most of the defects in the fabric surface with a clear orientation of the warp and weft, so in many cases the gray projection on the directions can enhance the abnormal values and highlight the defects. After the gray projection the image data becomes one-dimensional data. By studying the S transform coefficients characteristics of the normal points and defect points, give the rule of selecting the S transform coefficients, then combine the coefficients to remove the background and the noise, thereby enhancing defect. Use adaptive threshold for segmentation, then distinguish the presence or absence of defects. The fabric images with broken stitch are selected for the experiment and the simulation results show the effectiveness of the algorithm.(2) Propose a new algorithm for fabric defect segmentation based on the intrinsic frequency and one-dimensional S transform. By analyzing the S transform coefficients of the periodic pattern signal, introduce the concept of "intrinsic -frequency", which is incorporated into the fabric texture image analysis. According the difference in the intrinsic-frequency between the normal point and the defect point, determine the frequency threshold in S transform domain; remove the high-frequency coefficients; inverse S transform to achieve the purpose of removing the rule of texture. Furthermore by using of threshold segmentation, extract the defect information to complete the task of fabric defect detection. Finally, based on the defect area property get the final results that the presence or absence of defects. Experimental results show the algorithm is feasible and effective.(3) Present a new method for fabric image characteristic analysis based on 2D S transform and apply it to defect detection. Converting the two-dimensional image signal into a one-dimensional signal for S transform is feasible, but lack of correlation between pixels in adjacent rows. For the two-dimensional image signal, converts it into a one-dimensional signal S transformation is certainly feasible then, but considering the lack of correlation between adjacent rows of pixels. First, convert the fabric image by from the spatial domain to the S transform domain using the two-dimensional; remove of the rule of fabric texture and leave the defects and part of interference by setting the frequency threshold; then use threshold segmentation in the anti-S transform; finally, determine the presence of the defects. Experimental results demonstrate the feasibility and effectiveness of the algorithm.At last, summary the work on the text, review the main results of this study, and point out the direction for further development and research.