基于分数阶总变分和结构稀疏表示的图像去块算法

Image Deblocking Algorithm via Fractional-Order Total Variation and Structural Sparse Representation

摘要: 基于块离散余弦变换的编码技术，是图像与视频压缩的基石，但其固有的分块处理和量化步骤往往会引入诸如块效应、纹理退化及边缘模糊等多种失真现象，严重损害了图像的视觉质量。为解决上述问题，本文将分数阶总变分(FOTV)稀疏约束嵌入基于字典表示的图像块效应去除模型中，并通过优化分数阶阶数以自适应平衡去块强度与细节保持能力。该模型利用FOTV在刻画纹理细节方面的优势，在强力抑制块效应的同时，能更好地保持边缘结构与复杂纹理。实验结果表明，所提出方法在客观指标与主观视觉质量上均优于或达到现有多种去块算法的水平，特别是在保持边缘锐利性与纹理一致性方面展现出显著优势。

Abstract: Block Discrete Cosine Transform based coding technology serves as the cornerstone of image and video compression. However, its inherent block-based processing and quantization steps often introduce various distortions such as block artifacts, texture degradation, and edge blurring, which severely impair the visual quality of images. To address these issues, this paper introduces Fractional-Order Total Variation (FOTV) into the field of image deblocking and constructs a restoration model integrated with structural sparse priors. By optimizing the fractional order, the model adaptively balances deblocking strength and detail preservation capability. Leveraging the advantage of FOTV in characterizing texture details, the proposed model can effectively suppress block artifacts while better preserving edge structures and complex textures. Experimental results demonstrate that the proposed method outperforms or achieves comparable performance to several state-of-the-art deblocking algorithms in both objective metrics and subjective visual quality. In particular, it exhibits significant advantages in maintaining edge sharpness and texture consistency.