Research On Computer Vision Recognition Algorithm Based On Texture Image

Texture,main sources of visual perception for human,is a basic feature used to describe the surface of objects,which exists widely in nature.Texture analysis is an important research field in computer vision,meanwhile,it is an significant means of image processing.Thus,texture analysis is used in many fields.Although methods are similar for different problems,it is different that what places texture information extracted from pictures is applied in.Its applications include texture classification,texture segmentation,texture compression and texture synthesis.For texture classification,the biggest challenge is the extraction of texture features.In order to realize the result that the difference between classes is the smallest and the difference between classes is the biggest,how to extract feature becomes more and more count for different environment factors,such as light source and angles.Therefore,this thesis mainly studies texture feature extraction.The main contents can be summarized as follows:(1)At present,there are many algorithm to extract texture feature,however,they have a common problem,it is difficult to distinguish from different types of textures in various light sources and shooting angle.Here,we present a complete local binary pattern operator model of scalable block size(SB-CLBP,Scalable Block CLBP).At first,the algorithm extracts texture picture's CLBP feature.Secondly,By means of the fixed size sliding window,the texture map is processed by mean value,and then the CLBP feature is calculated again.Though the process,coarse grained information of texture is embodied.Finally,the coarse grained information of texture is added to original picture's CLBP.It is Equivalent to the combination coarse-grained information and local information combination,the ability to describe the texture is further enhanced.The method proposed in this paper can not only distinguish the texture images with great difference under different light sources and angles,but also can distinguish the categories with minor difference.(2)The feature extraction effect of the blurred image is poor,for this circumstances,this paper is not blindly do fuzzy operation,but improve the contrast between the texture by image enhancement to highlight its texture information?The algorithm first do bilateral filtering operation on blurred image,it can smooth the image while preserving the edges of the texture information,and then minus the original image and multiplied by magnification,finally add the original image to obtain the final enhanced image.Experiments on Outex and CUReT texture libraries show that the proposed SB-CLBP feature extraction algorithm improves the accuracy of texture classification effectively.In order to verify that the algorithm has good texture description ability,this paper applies it to the feature extraction of barcode,and proposes a barcode location algorithm based on SB-CLBP feature and SVM.Experiments on the barcode library WWU(Westfalische Wilhelms-Universitat Muenster Barcode Database)show that the proposed localization algorithm achieves satisfactory results,and also verifies that the SB-CLBP feature extraction algorithm is superior to other algorithms.