Abstract:

Image enhancement is essential pre-processing for many computer vision applications, especially for low-visibility scenes. This study develops a unified two-pathway model for image enhancement tasks like HDR image tone mapping and LDR image enhancement. Early visual mechanisms inspired this model. The structure-pathway and detail-pathway, which encode low- and high-frequency visual information, receive the input image in two parts. In the structure-pathway, an extended biological normalisation model integrates global and local luminance adaptation for visual scenes with different illumination levels. The detail-pathway enhances detail and suppresses noise using local energy weighting. Finally integrating structure- and detail-pathway outputs enhances low-light images. The model can be modified to tone map HDR images. The proposed model outperforms related state-of-the-art approaches and can handle the aforementioned visual enhancement tasks

Description:

night time image, low-light image, HDR image, visual adaptation, noise suppression.