Region Wise Surface Level Defect Detection and Ranking of Crust Leather Images Based on Image Processing Techniques

Abstract

Sorting and aligning of crust leather for grading on position wise defect distribution is one of the methods adopted in the tanning industry. This method is generally carried out manually by a veteran on official sampling position and their input is critical because it is directly linked to sales of the crust leather. The opinion of the experts is believed to be stable and consumes a good amount of time too. Hence, in the current research a robust defect detection method and ranking of crust leather images based on image processing techniques is proposed to give a stable solution in a short span of time. A custom-made dataset of crust leather images consisting of 5640 images were used in this study. The pixel intensity has been extracted on demarcated position of various regions including neck, belly left, belly right, center and butt instead of official sampling position through horizontal and vertical mapping of coordinates with a new method Grading Score on Image Position wise (GSIP) on the actual images. The image processing techniques using Canny Edge Detection and filters such as Laplacian, Median, Prewitt, Roberts, Sobel and Scharr were implemented to get the pixel intensity grouped and classified based on parameters within acceptable range using a Naïve Bayes Classifier. The classifier confirms that the accuracy of Set I - Actual Images and Set II - Defects with implementation of canny edge detection over other image processing techniques at 99.50%. Therefore, the current research confirms that the proposed GSIP method would give an additional tool to inspectors while ranking the crust leather based on region wise surface level defect detection of crust leather images based on image processing techniques.