The phase-based technique was shown to be a valuable method for disparity and optical flow estimation. The method is based on the convolution of images with Gabor filters. This article shows that the convolution operation is equivalent to the solution of a linear system of differential equations if the Gaussian kernel of the Gabor filters is replaced by different types of kernels. From a functional point of view, the system behavior can be considered as the steady-state solution of a diffusion process with lateral inhibition. The continuous model is extended to a discrete lattice model in which the degree of the differential equations is reflected in the number of interconnections between nodes required to perform the computation. A potential VLSI implementation of the algorithmic procedure in which interconnections between nodes are limited to the nearest and the next nearest neighbors is discussed
Continuous and Discrete Models for Phase-Based Stereo Vision
1998-01-01
Abstract
The phase-based technique was shown to be a valuable method for disparity and optical flow estimation. The method is based on the convolution of images with Gabor filters. This article shows that the convolution operation is equivalent to the solution of a linear system of differential equations if the Gaussian kernel of the Gabor filters is replaced by different types of kernels. From a functional point of view, the system behavior can be considered as the steady-state solution of a diffusion process with lateral inhibition. The continuous model is extended to a discrete lattice model in which the degree of the differential equations is reflected in the number of interconnections between nodes required to perform the computation. A potential VLSI implementation of the algorithmic procedure in which interconnections between nodes are limited to the nearest and the next nearest neighbors is discussedI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
