Fabric Defects Detection Algorithm Based On Haar Wavelet Decomposition And Gray Gradient Enhancement

As an important raw material for light industry,the quality of cloth has always been one of the core issues that various enterprises in the textile industry attach importance to.About years ago,due to the limitations of the technique,the defects detection of the cloth surface has always relied on the workers' eyes on the assembly line.There are some problems to detect the defects through this way,such as high error rates,difficult to unify standards,and human influence.With the development of technology in recent years,the automatic detection technology of cloth is more and more applied in the actual production of factories.It makes the standardization of the fabric industry to be possible,and also promotes the development of related industries.With the development of related technologies,many different algorithms based on machine vision have been proposed.These specific methods can be roughly divided into statistical methods,spectral methods,and model-based methods.This thesis has done the following work for the practical problems encountered in the detection of fabric defects:1.An enhanced non-local self-similarity algorithm for fabric defect detection.Aiming at the problem that the detection method based on image non-local self-similarity(NSS)cannot effectively detect small linear defects,We chose the enhancement method based on gray gradient to improve the NSS algorithm,which named as the enhanced non-local self-similarity defect detection method(ENSS).Experimental results show that the algorithm solves the problem that the original algorithm lacks detection capability for small linear defects while ensuring the original detection performance.2.A defect detection algorithm based on Haar wavelet decomposition and variance.Aiming at the problem that many current detection technologies cannot detect small defects or fuzzy defects effectively.We choose the wavelet-based method to design the fabric defect detection algorithm.First,we use a mean filter to pre-process the collected image,perform the Haar wavelet decomposition on thepre-processed image,and select the LH channel,that is,the image representing the vertical high-frequency coefficient for enhancement,and suppress the information of the other three channels.After that,the restructured image is divided into many equal-sized small blocks.We calculate the variance of each block,combined with the variance-based threshold method to determine whether there are defective areas in the small blocks,and then synthesize these blocks into a new image to achieve defect location.Experimental results prove the effectiveness of the algorithm.