A HYBRID SYSTEM BASED ON SAM-2 AND A LATENT CROSS-ATTENTION DIFFUSION MODEL FOR AUTOMATIC OBJECT SEGMENTATION AND CONTEXTUAL INPAINTING IN IMAGES

Abstract

This article addresses the problem of high-precision segmentation of objects in images and their contextually accurate restoration through inpainting. The authors propose a novel hybrid system based on the Segment Anything Model 2 (SAM-2) and a Latent Cross-Attention (LCA) diffusion mechanism. The SAM-2 model enables precise delineation of object boundaries in complex backgrounds, while the LCA-Inpaint module fills missing regions while preserving semantic consistency with the remaining parts of the image. Experimental results demonstrate that the proposed method achieves 15–20% higher performance than existing models in terms of PSNR and SSIM metrics.