Study On Woven Fabric Texture Based On Fourier Transform And Gabor Transform

Texture was the reflection of the object surface information through the forms of thedistribution of image color or grayscale. Woven fabric texture was periodic and complex becauseof the changing of yarn and interweaving pattern. Fabric texture analysis methods play a keyrole in the automatic inspection of fabric pilling, fabric drape rating, fabric defect detection andimage segmentation. In the above study the texture enhancement or elimination, the directionanalysis and the extraessential texture enhancement methods were basic and these generalmethods were important.The main work of this paper was as following:The woven fabric image was converted from the spatial domain to the frequency domainusing Fourier transform, then the spectrum of the woven fabric was obtained. The appropriatethreshold value was selected from the threshold segmentation curve. Then the spectrumdiagrams were divided with the threshold value and the feature points containing the main fabrictexture were found by the method with maximum of the neighborhood. The frequency range ofwoven fabric texture was from2to8cycle/mm based on the relationship between the prominentpoints and the real frequency. The influence on fabric frequency of image resolution and fabricdensity was analyzed.The main parameters of Gabor filter was determined by the woven fabric texture frequency,which was obtained by Fourier transform. The filter size, center frequency, decomposition scalesand directions were discussed. Then the woven fabric texture was filtered by Gabor transform.Various sub-images with different information were obtained after the woven fabric imagedecomposition by Gabor filter along different scales and directions. Plain, twill and satin fabricimages were transformed using2to5scales,3to6directions and the fabric texture frequencywas normalized from0to1respectively.The sub-images were fused to obtain specific texture information while three fusionmethods were used and compared. The specific texture information can be acquired by differentfusion strategies including fusing the sub-images from single scale, single direction or selectingsub-images with a maximum feature. Even the sub-images could be selected optionally. The suitable fusion method, which has obvious effect on woven fabric texture analysis, was chosenaccording to the fusion evaluation index.Woven fabric was formed with different warp and weft yarns according to certain rules. Themain fabric texture usually presents with a certain angle. Fabric texture also contains someextraessential texture, which was formed with defects, noise, luminance unevenness or otherreasons. The woven fabric texture was analyzed with the plain, twill and satin fabric bytransformation with FFT and their frequency were analysed. Then the images were decomposedwith Gabor filters and the specific fabric texture could be obtained by fusing several selectedsub-images. Finally, the main texture enhancement method, the main texture elimination method,the direntional texture analysis method and extraessential texture enhancement method werediscussed.The experiment results showed that the enhanced fabric texture can be obtained from thesub-image which contains the periodic texture. The abnormal texture such as defects, pillingsand wrinkles could be obtained by eliminaling the main texture or enhencing the extraessentialtexture. This woven fabric texture analysis method, which was based on Fourier transform andGabor transform, could obtain woven texture image with different frequency and directions, andwould be a powerful tool in the application of texture analysis.