A simple computational framework for defect detection system with orthogonal polynomials transcoded coefficients

In this paper, a simple computational framework to propose a generic defect detection system based on orthogonal polynomials transcoded coefficients, with a statistical procedure is presented. Initially, the defective input image is partitioned into blocks and subjected to orthogonal polynomials transformation. The resulting coefficients are then applied with a modified lifting scheme, to produce transcoded coefficients with reduced block size. These coefficients are modeled as a probability distribution to propose a block classification scheme that classifies the block under investigation to have dominantly either texture or edge or smooth with total number of transcoded coefficients that are above the mean of the sample. With simple statistical procedure, we then introduce a new defect detection technique on each of these block classification result. The proposed defect detection technique employs homogeneity among variance of transcoded coefficients with Box's M Test, and group distribution model to verify the presence of defect in texture and edge block respectively. By analyzing the magnitude of transcoded coefficients, defective blocks in a smooth region are identified. The proposed defect detection system, an application independent, is experimented with natural images and few measures are introduced with a simple Defect Measurement Matrix (DMM) to analyze the performance of the proposed system. The applicability of the proposed scheme is also extended to identify defects in fabrics.