PTCTV-KMC: Infrared Small Target Detection Using Joint Partial Tensor Correlated Total Variation and K-Means Clustering

Abstract

Infrared (IR) small target detection is widely utilized in both military and civilian sectors. Despite the development of numerous advanced methods for detecting small targets, improving overall performance in complex scenes remains a significant challenge. To address this issue, we propose a joint partial tensor correlated total variation and k -means clustering (PTCTV-KMC) method that integrates local and global features. The proposed method comprises two stages: seed point (i.e., candidate target) search and seed point discrimination. In the seed point search stage, a tensor low-rank sparse decomposition model is first used to decompose the IR image into target and background images. To reduce residual noise and background edges in the target image, we designed a partial tensor correlated total variation (PTCTV) norm. This norm effectively constrains the global low-rankness and local smoothness of the background, and enhances the model's focus on image detail information. Subsequently, leveraging the global sparsity of the target, a density peak search technique is employed to locate seed points in the target image. In the seed point discrimination stage, k-means clustering is utilized to improve the accuracy of the local contrast measure (LCM) in scenarios with uncertain target distribution and mixed background components. By calculating the LCM for each seed point, we further suppress background clutter and enhance real targets. Extensive experiments demonstrate that the proposed method exhibits superior overall performance compared to advanced methods and achieves satisfactory computational speed.