Image Edge Detection Method Based On Anisotropic Gaussian Filtering

Edges are one of the most important basic features in images and convey most information in images. The information is of great value for people on high-level image processing, such as feature extraction, object identification, image segmentation, and image understanding. Edge detection is an indispensable pre-processing of sequent high-level image processing task and the quality of edge extraction directly influences the overall performance of sequent high-level image processing systems.The edges in an image indicate the pixels where the gray-level of the image have sharp changes and edge pixels often consist of the boundaries of objects of interest in the scene and contain much information. In most cases, images are corrupted by noise. It is an active area how to extract efficiently the edges from noisy images. Gaussian filters and their derivatives play an important role in traditional edge detection methods. Images are first smoothed by a low pass Gaussian filter to suppress noise and then the first-order or second-order derivatives of the smoothed images are used to detect edge pixels. When isotropic Gaussian filters are used, the detectors suffer from an unavoidable conflict between noise suppression and edge blurring. But traditional edge detectors are all based on isotropic filters which can not make good use of edge's orientation information. And when edges locate very near with each other, isotropic Gaussian filtering will blur the edges, thus can not achieve good edge detection results. In this thesis, we investigate the edge detection method using anisotropic Gaussian filters and anisotropic directional derivatives. In the proposed method, a noisy image is first smoothed by a set of anisotropic Gaussian filter with a same scale and different directions. Then, the directional derivatives from different smoothed versions are calculated. Finally, the maximal directional derivatives are used to detect edges in the Canny framework. The proposed detector can better extract edge from noisy images.Besides, we also apply the proposed detector to color RGB images for edge detection. The detector is applied to the R, G, and B components of the image to extracted three edge maps. Then, the three edge maps are fused into an edge map. The experimental results show that this method can obtain better detection performance.