Practical Contour Tracing Via Integration of Adaptive Contrast Stretching and Gabor Wavelet Transform

Abstract

The Gabor filter is applied using discrete wavelet transform for edge detection and contour tracing on true color images, whose wavelet functions provide the high resolution in both spatial and frequency domains. The number of wavelets also affects the quality of edge detection. To generate more accurate outcomes, integration of adaptive contrast stretching and Gabor wavelet transform is properly carried out for edge detection and contour tracing on broadly selected images. Contrast stretching is based on adaptive histogram equalization generalization, where the accumulation function is used to generate intensity mapping of three primary color components stem from local histograms. Any digital image consists of major objects, edges, corners and blobs as well as the noises and background fluorescence. Critical changes in image properties can be captured via detecting sharp variations in the intensity. Although the gray level intensity is commonly used for edge detection, the mixture of true color components (red, green, and blue) gives rise to an overall appeal with better visual effect and color balance by rendering three specific colors. In fact, edges, contours and boundaries could be detected in various ways by means of intensity changes, however, edge broken and false detection are among typical drawbacks in classical approaches, which are subject to information loss and feature deformity. The popular Canny edge detection (CED) and Ant Colony Optimization (ACO) detection are among the most effective approaches for edge detection and contour tracing. With the benefit of artificial intelligence, the performance of ACO can be slightly better than CED, however, the tradeoff is the much larger computation complexity. Instead, Gabor wavelet detection offers similar computation cost to CED, but dramatically better outcomes are produced with relatively less efforts. A comparison between Canny edge and Gabor wavelet detection is also made in terms of both visual appealing and quantitative evaluation.