Studies on the development of image processing methods for leather species identification defect detection and grading

Abstract

Leather surface images provide useful information about surface morphological features as well as defects. Identifying species and detecting surface defects using leather images are of great scientific interest in the objective assessment of quality. The developments in image processing techniques and computing power provide avenues for development of automated techniques for feature extraction from leather surface data. Leather being a natural material, surface quality of the leather varies quite considerably due to the inherent variation in the raw material exhibiting uneven surface characteristics, low contrast etc. This poses challenges in detecting/mapping morphological and defect features from leather images. Identifying these features manually is a time-consuming endeavour. A reliable objective automated detection approach is sought after to replace conventional manual methods which are subjective, inconsistent and laborious. This thesis presents a research endeavour whose aim is: (i) to design and develop a robust machine learning algorithm for leather species classification (ii) to design and develop a prototype to acquire images of surface defects over the entire area of the leather (iii) to develop texture feature descriptors to discriminate defective and non-defective regions of the leather and to develop grading rules based on computation of effective usable area of the leathers for objective leather grading. Leather species identification refers to finding the animal from which the hide or skin is obtained for making leather. An efficient and objective method for species identification would provide product authenticity by prevention of product counterfeiting thereby helping in setting trading standards. It will also be useful in the prevention of use of skins from endangered species. A unique non-destructive leather species identification algorithm has been proposed for the identification of cow, buffalo, goat and sheep leathers. Hair pore pattern was segmented efficiently using Fast Convergence Particle Swarm Optimization algorithm. Significant features representing the unique characteristics of each species such as number of hair pores, pore density, percent porosity, shape of the pores etc., were extracted. Experimental results on the leather species image library database achieved a prediction accuracy of ~90 % using multilayer perceptron neural network as classifier, confirming the potentials of using the proposed system for automatic leather species classification. newline