Partition-Based Image Exposure Correction via Wavelet-Based High Frequency Restoration

In the field of image processing, exposure is often a crucial factor influencing image quality. Existing image enhancement methods typically focus on addressing a single exposure issue either under-exposure or over-exposure. Most methods simultaneously addressing multiple exposure issues do not work well, and their effectiveness is also limited. In response to the above challenges, a novel exposure correction method has been proposed. Firstly, utilizing the structural sensitivity of U-shaped network, an Illumination Attention Map Estimation Network (IAMEN) is designed to estimate the partitions of an image. Secondly, a Partition-based Enhancement and Refinement Network (PERN) is proposed. In the enhancement stage, the illumination attention map obtained by IAMEN guides PERN to focus on different exposure areas. The two branches of Partition-based Convolution Enhancement Module (PCEM) incorporate an illumination attention map and its supplement to 1, allowing them to focus more on the under- and over-exposure areas, respectively. The encoder in PERN consists of three PCEMs. In the refinement stage, a High-frequency Feature Refinement Module (HFRM) is proposed to extract image high-frequency features using discrete wavelet transform for edge enhancement. Extensive experiments demonstrate that the proposed method consistently achieves remarkable performance compared to several state-of-the-art methods.