A New Multi-scale/multi-directional Method for Detection of Abnormalities in Random Textures

This paper proposes a new Gabor filter-based generative approach to texture abnormality detection, called Gabor Composition or GC. The GC algorithm, a combination of Gabor filtering and co-occurrence analysis, employs a Gabor filter bank to generate a multi-scale and multidirectional feature map of the test image. The feature map is then fed to a co-occurrence based feature extraction module, which is able to extract more efficient textural features from the conditioned signal. A comparative defect detection test applied on a database of randomly textured ceramic tiles, illustrates the advantages of the proposed method.