Machine Vision Inspection System for Detection of Leather Surface Defects

Abstract

Leather quality inspection is very important in assessing the effective cutting value that can be obtained from the leather. Current practice involves an expert to inspect each piece ofleather individually and detect defects manually. However, sucha manual inspection is highly subjective and varies quiteconsiderably from one assessor to another. Often this subjectivityleads to dispute between the buyer and the seller of the leathersand hence attempts are made to automate this. Automatic leatherdefect classification is a challenging research problem due to thedifficulties that arise when segmenting defects from the leatherbackground and determining the characteristics that describethe defects objectively. The present study describes application ofmachine vision system to capture leather surface images and thenovel multi-level thresholding algorithm to segment defectiveand non-defective regions of leather followed by texture featureextraction to objectively quantify the leather surface defects. Adataset consisting of 90 leather images comprising 20 goodleather and 50 defective samples has been used in the study.Experimental results on the leather defect image library databaseachieved an accuracy of 90% using neural network as classifier,confirming potential of using the proposed system for automaticleather defect classification.