Robust Defect Detection on 3D Textured Low-Contrast Surfaces Using Accumulated Aggregation Shifting Model

Detecting defects on 3D textured low-contrast surfaces plays an important role in product quality control. However, because of affects from the uneven distributions of material, irregular textures, and unclear boundary between defect and background, this is a very challenging problem. In this paper, we propose an unsupervised defect detection method guided by saliency. Firstly, two features, named local-global intensity difference and local intensity aggregation, are proposed to measure saliency of each pixel. These two features are further utilized to construct an accumulated aggregation shifting (AAS) model, which iteratively shifts brightness of pixels based on their visual saliency, i.e. defective probability. And then, the output sequence of AAS at different iterations can be formalized as linear distribution or exponential distribution through statistical analysis. Finally, by utilizing the risk minimization method, we theoretically determine a reasonable threshold to classify all pixels as defective ones or defect-free ones. Experiments on a real industrial image dataset demonstrate the effectiveness of our approach.