Image Analysis through the Detrending Moving Average Algorithm

Texture classi cation is an important rst step in image segmentation and image recognition. In this report we present the Detrending Moving Average (DMA) algorithm as a robust and informative classi cation algorithm. We train the DMA algorithm with the images of the UIUC dataset. The Cholensky-Levinson Factorization algorithm is used to generate arti cial fractal surfaces as a reference dataset. In the classi cation results the DMA algorithm seems to be able to detect scale, aspect and rotation changes in the analysed random textures.