A Fast Sparse Reconstruction Approach for High Resolution Image-based Object Surface Anomaly Detection

We propose an approach to resolve two issues in a recent proposed sparse reconstruction based, anomaly detection approach as a part of automated visual inspection (AVI). The original approach needs large computation and memory for high resolution problem. To solve it, we proposed a two-step sparse reconstruction, 1) the first sparse representation of input image is estimated in a sparse reconstruction with low resolution downsampled images and 2) the high resolution residual values is generated in another sparse reconstruction with the sparse representation. The first step provides the flexibility of freely adjusting the computation and the demand of memory storage with small trade-off of detection accuracy. Moreover, an illumination adaptive threshold with morphological operators is used in the anomaly classification. Empirical results show that the proposed approach can effectively replace the original approach with better results.