Texture Image Recognition Using Bispectrum Slice

This paper presents a novel texture recognition method using bispectrum slice. The first step, Radon transform, was to reduce the dimension of the image data. The second step was to calculate bispectrum and extract bispectrum diagonal slices as texture features. The third step was to apply principal component analysis(PCA) for reducing the dimension of feature vectors. Finally, BP(Back Propagation) neural network based on resilient BP algorithm was used as training and classification. The results show the bispectrum slice is more successful than gray level co-occurrence matrix(GLCM), and has a recognition ratio of 87.33%. The bispectrum-based approach can effectively recognize different texture images and sufficient texture information can be obtained.