Divide-and-conquer framework for image restoration and enhancement

Abstract

We develop a novel divide-and-conquer framework for image restoration and enhancement based on their task-driven requirements, which takes advantage of visual importance differences of image contents (i.e., noise versus image, edge-based structures versus smoothing areas, high-frequency versus low-frequency components) and sparse prior differences of image contents for performance improvements. The proposed framework is efficient in implementation of decomposition-processing-integration. An observed image is first decomposed into different subspaces based on considering visual importance of different subspaces and exploiting their prior differences. Different models are separately established for image subspace restoration and enhancement, and existing image restoration and enhancement methods are utilized to deal with them effectively. Then a simple but effective fusion scheme with different weights is used to integrate the post-processed subspaces for the final reconstructed image. Final experimental results demonstrate that the proposed divide-and-conquer framework outperforms several restoration and enhancement algorithms in both subjective results and objective assessments. The performance improvements of image restoration and enhancement can be yielded by using the proposed divide-and-conquer strategy, which greatly benefits in terms of mixed Gaussian and salt-and-pepper noise removal, non-blind deconvolution, and image enhancement. In addition, our divide-and-conquer framework can be simply extensible to other restoration and enhancement algorithms, and can be a new way to promote their performances for image restoration and enhancement.