Fast defect detection algorithm on the variety surface with random forest using GPUs

This paper proposes a defect detection method that can find the surface irregularity of the variety surface. The human like, the Variance of Variance (VOV) method can calculate the characteristic pattern of surface with mathematization. Also defect with unusual pattern appear using the VOV method. The object surface usually has smooth or textured plane, including defects with intensive irregularity. Conventional variance detection algorithms are not adequate for detecting various defect types of multiple scales and intensity variations. The purpose of this paper is to propose a method that can find defections of different scales in a single framework. This method can detect defects of different sizes by changing the window´s size, which is combined with GPU computing for real time processing of inspection. The local VOV value does not affect other calculation result that can good for parallel processing. GPU has the high performance with parallel processing and it solved the typical processing speed problem of VOV method. For robust detection, we combined result of single VOV detection using machine learning process. From the results of the experiments that used in real images, it is verified that the method can be applied to detect the defects of images for irregular and textured intensity under uneven illumination.