Identification of surfaces using discrete triangular approximation of Gaussian curvature

Abstract

Object recognition is one of the prime probems of computer vision. One way of extracting information is to compute the Gaussian curvature for the given surfaces. The algorithm uses discrete approximation using triangularization methods to compute Gaussian curvature. The images are initially broken down into different segments and the Gaussian curvature for each pixel in the segment is computed with respect to its eight neighboring pixels. These computed values are then converted into intensity format for graphical visualization. The images with improved edge information have been taken from previous work. Synthetic images containing signal object scenes have been tested.