Research On Fabric Texture Recognition Method

Research on automatic recognition of woven fabric pattern not only plays an important role on the application of recognition for woven fabric pattern,but also is a strong representation for study of texture analysis and recognition algorithm.In this paper we improve the flow and framework of woven fabric pattern recognition method,after studying deeply on texture analysis and recognition based on related knowledges.A two-step method for woven fabric pattern recognition is proposed.The first step is woven fabric pattern classification based on yarn boundary feature.The second step is the final recognition of woven fabric pattern based S-Gabor feature and differentcategories-correction method.The two-step method integrates two features at different operation as they have unequal functions during the whole process.When calculating the yarn boundary feature representing crossed-area of the warp and weft yarn,a yarn boundary information identification method based on luminance variation is proposed.In this method,after local crossed-area normalization process,each crossedarea's yarn boundary feature can be got from two different luminance variation relationships between neighbouring crossed-areas,where one relationship is called ‘Absolute luminance variation relationships between neighbouring crossed-areas' and the other one is called ‘Relative luminance variation relationships between neighbouring crossed-areas' in this paper.When calculating the size of the fabric circular unit,average Hamming distance is used to indicate the similarity between two yarns.And then using average correlation coefficient between a group isometric subsequence,which is segemented from the similarity sequence with different length parameter,to correct the size of the fabric circular unit,which has been calculated based on the minimum points of the similarity sequence.Then a woven fabric pattern classification scheme based the size of the fabric circular unit and the oblique correlation of the pre-recognition result of woven fabric pattern is proposed.A kind of wavelet transform called S-Gabor transform derived from Steerable Filter and Gabor transform is proposed in this paper for accurate recognition of woven fabric pattern.This paper then analyses the properties of the S-Gabor transform and explains the visual principles of this transform.The S-Gabor transform is verified to have an advantage in image gradient information extraction in theory.Further,S-Gabor transform is used to extract features of each crossed-area in non-twill weave and then reduce the dimensionality of these features based on PCA.After the dimensionality reduction operation,the crossed-areas in non-twill weave is divided into two categories,warp cross point and weft cross point,using SVM.After all,different kind of error recognition correction method is proposed to correct the error in pre-recognition of plain weave,twill weave and variety twill weave or satin weave innovatively.The woven fabrics image database used in this paper came from Tianjin Polytechnic University.As having characteristics include of comprehensive woven fabric pattern species,changing image information and complex interfering factors,this database is of value to study and reference.Results of the experiment in this paper proved that the S-Gabor transform can extract the usefull texture feature effectively in woven fabric pattern recognition.The metod of fabric circular unit size calculation,woven fabric pattern classification and woven fabric pattern recognition proposed in this paper has very high accuracy rate and robustness in the woven fabric sample gallery used in this paper,which can reach 99.25%,99.62%,96.98% respectively.The woven fabric pattern recognition proposed in this paper can balance the high efficiency of recognition method based on yarn boundary feature and high accuracy of recognition method based S-Gabor features well.