| With the rapid development of "Internet +" and computer science,various industries are increasingly demanding real-time calculation of massive data,especially in the field of high-resolution images,high-definition image calculations,and real-time online detection of images.The calculation of an image often needs to calculate each pixel in the image.In many fields,some operations need to be used to preprocess the image before the next operation can be performed.The processing efficiency of preprocessing often determines the entire image processing.The processing efficiency of the process is even more significant in the field of video processing,because the processing efficiency of the entire video is determined by the processing efficiency of a single picture.Therefore,it has become a key issue to improve the traditional algorithm to use the parallel method as much as possible and optimize the original algorithm.The textile industry is also developing rapidly in the direction of computer assistance.The automation of the weaving industry is not simply the automation of weaving equipment.It is also reflected in the quality control of fabric products,which includes many aspects such as fabric design,testing and classification.With the advancement of various types of image acquisition technology and the rapid development of image processing theory,product design,pattern recognition,and quality monitoring of woven fabrics can use computer image processing technology to assist traditional manual methods.For each link in the fabric production process,the main research contents of this article can be as follows:(1)An improved Kirsch operator for fabric pattern contour extraction under CUDA is proposed.From the algorithm optimization level,the traditional Kirsch operator’s edge detection algorithm is reduced.From the parallelization acceleration level,the twodimensional point set is designed according to the essence of the image,and a secondary threading mode is designed.After testing,compared with the traditional Kirsch operator’s edge detection algorithm,the proposed algorithm has higher execution efficiency;pictures of different scales have a gradually increasing and flattening trend on the parallel acceleration ratio,and their acceleration efficiency and image scale Size related.(2)Aiming at the problem that the obvious features of the fabric defect image with background pattern make the target defect detection difficult,a smooth amplitude spectrum is used to filter the frequency-domain information to reduce the effect of printing background on fabric defect detection.In view of the limitations of global operations that directly perform a univariate function transformation on the amplitude,a multivariate local smoothing strategy is proposed to suppress repeated patterns in the image.According to the feature maps obtained at multiple smooth scales,the feature map with the smallest information entropy is selected as the final feature map,and the frequency domain features are extracted using scale filtering.The data test shows that compared with the univariate function transformation method,this method can effectively suppress the background of complex repeated fabric patterns,and the flaw detection has a better effect.(3)A fabric defect detection method based on Elo grade system is proposed.This method is based on sports competition to achieve defect detection,that is,fair matching competition between image partitions.The image can be divided into several standard size images.By using the initial reference score,a competition is performed among the numerous partitions to update the Elo rating matrix.The bright flawed areas are regarded as strong athletes who often win the game,the flawless areas are regarded as normal athletes,the competition results are more average,and the dark flawed areas are regarded as the weaker athletes who often lose the game.After all competitions are over,the brightly-defected areas will get a higher score,otherwise,the dark-defective areas will get a lower score.Experiments show that this method has a high success rate and a good detection effect. |
