FAST IMAGE RESTORATION AND ENHANCEMENT

Computer Engineering

This paper presents MIRNet-v2, a novel architecture for image restoration that aims to preserve high-resolution spatial details while effectively leveraging contextual information from low-resolution representations. The proposed approach utilizes multi-scale residual blocks with parallel multi-resolution convolution streams, mechanisms for information exchange, non-local attention mechanisms, and attention-based multi-scale feature aggregation. MIRNet-v2 achieves state-of-the-art results across various image processing tasks, including defocus deblurring, image denoising, super-resolution, and image enhancement, as demonstrated through extensive experiments on six real image benchmark datasets. Experiments show that in deep learning, using image enhancement algorithms may improve CNN performance when training complete CNN models, but not all image enhancement algorithms can improve CNN performance; in transfer learning, when fine-tuning the pre- trained CNN model, image enhancement algorithms may reduce the performance of CNN.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.