Identification of Camouflaged Defects Through Central Moments And Hierarchical Segmentation

Abstract

Defects in some industrial products such as tiles, plates etc, generally go unrecognised by human vision based quality control and even by conventional instrumentation systems. This happens because the defective portions get blended with the background so much that the uniformity and continuity in texture appear to be maintained, in other words defects are said to have been camouflaged.
One alternative to detect such defects is to resort to more detailed texture analysis. In this paper, we propose to split the entire region into smaller blocks, and employ in each block, an advanced texture analysis model such as invariant central moments. Relatively, the camouflaged texture should exhibit a contrast in smaller blocks because local features of that block are indented in computing invariant central moments. Finally defective portions are identified by hierarchical segmentation model. Several experiments have been conducted to demonstrate the suitability of the proposed model to spot camouflaged defects.