Dense disparity estimation based on the bi-dimensional empirical mode decomposition and Riesz transformation

摘要: This paper proposed to apply the Bi-dimensional Empirical Mode Decomposition (BEMD) dense disparity estimation problem. The BEMD is a fully data-driven method and does not need predetermined filter wavelet functions. It locally adaptive has obvious advantages in analyzing non-linear non-stationary signals. Firstly we decompose original stereo images by 2D-sifting process of respectively. Through this procedure, serial Intrinsic Functions (IMFs) residue are achieved. denotes DC component signal. Secondly, subtract from image. resulting two dimensional signals can be thought being free disturbing frequencies, such as noise illumination components. Subsequently, obtain robust local structure information images, plural Riesz transformation utilized achieve corresponding 2D analytic images. Thirdly, extract phase similarity instead intensity information, taken basis calculating matching cost, which could reveal with more robustness. At last, map estimated based on method. winnertakes-all (WTA) strategy applied compute each pixel separately. Comparative experiment conducted compare performance intensity-based methods. Rather good results have been