Mltiscale Wavelet Representation And Imaginary Gabor Function Based Fabric Defect Detecction Algorithm Research

In fabric industry，the fabric defect detection is mostly finished by human, whichcould not ensure accuracy and high efficiency because of the heavy labor intensity,complex environment and other factors, and is very easy to cause damage to eyes.Hence, the use of automatic detection system to reduce costs and improve efficiency is areasonable choice.In this paper, three kinds of the detection algorithms has been studied, and two ofthem are realized through software, which are Wavelet based and imaginary Gaborfunction based, separately.To the blob defects, a multi-scale wavelet representation (MSWAR) based algorithmare studied in this paper. In this algorithm the edges produced by wavelet decompositionare united. The local roughness and global image homogeneity are calculated on thefusion signal generated by the united edges. To reduce the effect of uneven illumination,the Ostu binarization is adopted. This algorithm can detect almost all the blob defects,and partly detect the horizontal and vertical through defects, and have a low alarm rate.To the defects which run through the whole image in the horizontal or verticaldirection, this paper studied a imaginary Gabor function (IGF) based algorithm. In thisalgorithm,the horizontal projection signal and vertical projection signal are generatedby summing up image pixel values along columns and rows, respectively. Afterconvolve the projection signals with the coefficients generated by the imaginary Gaborfunction, the detect location will be found. Through this algorithm the calculatedamount caused by multi-scale and multi-channel of2-D Gabor filters are took out andalmost all of the defects in horizontal and vertical are detected.Another way called singular value decomposition (SVD) has also be studied in thispaper too. A picture is considered to be a matrix, and use the singular value of thematrix as the feature for the fabric defect detection.