Chapter Four - Robustness Analysis of the Reduced Fuzzy Texture Spectrum and its Performance on Noisy Images

Abstract

Abstract In a previous study, we introduced the fuzzy texture spectrum (FTS) as a texture spectrum (TS)–based method, where a given texture was characterized by its associated spectrum using fuzzy techniques. It allowed for the definition of a spectrum that was more likely to be visible to human perception. Afterward, with the aim of improving the computational efficiency of the FTS, we modified it by grouping in the same class all the texture units differing by rotations of 45 degrees. In the original TS encoding, it was suggested to use a 20 × 20 pixels window to characterize the texture of an image through the TS. We present a set of experiments for determining the size that is best suited for characterizing natural images using a database of Brodatz images. A second set of experiments is presented for analyzing the performance of the texture and fuzzy texture encodings against noise. Finally, we study the ability of both encodings to identify Brodatz images belonging to the same class. To carry out the experiments, we use information theory and similarity and dissimilarity measures. The results obtained have proven that, in general, fuzzy encoding outperforms the performance of the original encoding.