Adaptive Target Boundary Extraction in Natural Scene Using AGADMM-based Active Contour Model

Abstract

Snakes, or active contours, are used extensively in computer vision and image processing applications, particularly to extract object boundaries. However, problems associated with automatic initialization approach for parametric active models and Low SNR (Signal Noise Ratio) of small target in natural scene, which hinder the idea of applying active contours to segment small target in automatic recognition system. To solve this kind of problem, this paper presents a novel AGADMM (Average Gray Absolute Different Maximum Map)-based active model to extract small target boundary in natural scene. AGADM (Average Gray Absolute Different Maximum), the dissimilarity measurement between targets and background region, is used to enhance small target in natural scene; In AGADMM, the initial snake can be drawn automatically by threshold segmentation. Validity and real-time of the proposed algorithm have been shown in experiments.