Research On Contour Grouping For Low-light Images

Extracting the contour of desired structure from the noisy image is a fundamental problem in computer vision. To solve this problem, researchers have proposed the theory of contour grouping. Contour grouping is used to identify desired structure from noisy image, and very important to many advanced visual problems, such as target recognition and content-based image retrieval. With the development of night vision technology, it is applied more widely. Applying contour grouping technology to low-light image is of great practical significance.By using computer image processing technology and computer vision technology, this paper designs and implements a contour grouping algorithm under low-light condition, including denoising low-light images by wavelet domain Hidden Markov Tree (HMT) model, extracting edge from low-light images by using multi-scale theory, constructing grouping elements, selecting the appropriate Gestalt rules for modeling, extracting the contour of desired structure from low-light images.The main task of the paper includes:1. To research the characteristic of low-light images, and according to the need of edge detection and contour grouping, propose a low-light image enhancement algorithm based on local window variance and a denoising algorithm based on wavelet domain Hidden Markov Model(HMT);2. To research the image edge extraction, propose a wavelet multi-scale adaptive threshold edge detection algorithm for low-light images;3. To research the grouping algorithm based on the Gestalt rules, propose a grouping element construction algorithm to provide input for grouping, propose and implement contour grouping algorithm for low-light images, and improve the algorithm through the experimental analysis.Experimental results show that the proposed grouping algorithm can achieve better results in low-light images.