Modified Graph Laplacian Model With Local Contrast and Consistency Constraint for Small Target Detection

Abstract

The traditional graph Laplacian model has been widely used in many computer vision tasks. The small target detection technique is one of the most challenging computer vision tasks in various practical applications. This article presents a small target detection method by developing a modified graph Laplacian model with additional constraints. The proposed method is designed based on specific characteristics of small target: Global rarity, local contrast, and contrast consistency. First, we analyze the primal graph Laplacian model, and exploit its ability to describe global rarity, and isolate outliers from nonoutliers. Next, indicators measuring local contrast and contrast consistency are constructed to delineate local characteristics of small targets. Then, we integrate the indicators with the primal graph Laplacian model, and propose a modified graph Laplacian model for small target detection. In the confidence maps obtained by the proposed model, small targets are well enhanced, while backgrounds are significantly suppressed. Finally, a small target detection method is proposed based on the graph model. Extensive experiments on various real datasets demonstrate the effectiveness and superiority of the proposed method in detecting small targets.

Keywords

• Contrast consistency indicator
• graph Laplacian (GL) model
• local contrast indicator
• small target detection