Advances in Rotation-Invariant Texture Analysis

Alfonso Estudillo-Romero, Boris Escalante-Ramírez

Abstract: Robust rotation invariance has been a matter of great interest in many applications which use low-level features such as textures. In this paper, we propose a method to analyze and capture visual patterns from textures regardless their orientation. In order to achieve rotation invariance, visual texture patterns are locally described as one-dimensional patterns by appropriately steering the Cartesian Hermite coefficients. Experiments with two datasets from the Brodatz album were performed to evaluate orientation invariance. High average precision and recall rates were achieved by the proposed method.