PROJECT TITLE :

A Biological Vision Inspired Framework for Image Enhancement in Poor Visibility Conditions

ABSTRACT:

For many computer vision applications, image augmentation is an essential pre-processing step, especially for scenes with low visibility. Based on the principles of biological vision, particularly the early visual mechanisms, we have developed a two-pathway model that may be used to improve low dynamic range (LDR) image enhancement and high dynamic range (HDR) tone mapping. Structure-pathway and detail-pathway, which correspond to M and P routes, respectively, are transmitted from the input image into these two visual pathways, which encode low- and high-frequency visual information in the early visual system. In order to integrate the global and local luminance adaption into the structure-pathway, an expanded biological normalisation model is utilised. This model can handle visual scenes with fluctuating illumination levels. On the other hand, in the detail path, local energy weighting is used to enhance detail and reduce local noise. Finally, the low-light image augmentation is achieved by integrating the outputs of the structure and detail pathways. With some fine-tuning, the proposed model can also be used for HDR tone mapping. On three datasets (two LDR image datasets and one HDR scene dataset), extensive experiments show that the proposed model is efficient and outperforms the relevant state-of-the-art methods in terms of the visual enhancement tasks.