Typical Infrared Object Segmentation Based on Sparse Shape Prior and Variational Regularization

The infrared images captured by the uncooled detector often exhibit interference issues, such as blurred edge details and uneven grayscale distribution, which can significantly impact the accuracy of object segmentation. To address this, we propose an enhanced implicit shape representation framework based on a sparse representation model. This framework guides the evolution of implicit shapes using sparse linear combinations of probabilistic shapes drawn from a predefined dictionary. First, representative shape components are selected from the dictionary to form sparse combinations that effectively model the target shape. The object contour prior is implicitly incorporated into the sparse representation, facilitating more accurate contour alignment. A new energy function is then constructed, integrating region-based segmentation with sparse representation. The optimal level-set function is obtained through iterative optimization, ultimately yielding precise object segmentation results. Experimental evaluations demonstrate that the proposed model delivers robust segmentation performance, especially for typical objects in complex backgrounds.