Leather texture classification using wavelet feature extraction technique

This paper describes the application of image processing techniques for identification of leather defects using wavelet feature extraction method. Defects occur in leather and are identified at different stages of processing such as pre tanning, post tanning, crust, finishing etc., Manual defect inspection and analysis varies from person to person and is labor intensive, tedious, misjudgment occurs due to fatigue etc., As a consequence, the identification of leather defects becomes ambiguous that affects the quality control clearance of global trading and thus reducing the productivity. The objective of this research work is to identify leather defects from leather image library using wavelet based feature extraction technique. Captured leather images were processed in frequency domain with the help of using wavelet transform and were stored in the leather image library. Frequency component of an image can be analyzed better than its pixel intensities of an image because edges and uncorrelated pixels were well projected in frequencies. In wavelet transform, the frequency components of image was organized such that the lower and higher frequencies were separated, it also gives the image variations at different scales because of its Multiresolution analysis and hence makes wavelet more suitable for leather defect identification. The leather defects were identified by its texture using wavelet statistical features and wavelet co-occurrences matrix features such as Entropy, Energy, Contrast, Correlation, Cluster Prominence Standard Deviation, Mean, and local homogeneity. The classification of leather defects was done by employing Support Vector Machine (SVM) Algorithm with wavelet based feature extraction technique. Comparative analysis of different kernels used in SVM classifier was also discussed. Result suggests that the identification of leather defects can be automated with the application of image processing technique using SVM Classifier with wavelet feature extraction technique.