A BIOLOGICAL VISION INSPIRED FRAMEWORK FOR IMAGE ENHANCEMENT IN POOR VISIBILITY CONDITIONS USING ANISOTROPIC DIFFUSION FILTER

Abstract

Image enhancement with edge retention in the presence of outliers and artefacts is a daunting task in the field of image processing and computer vi sion. Inspired by the information processing mechanism in the human retina, an effective two-pathway image enhancement framework that incorporates anisotropic diffusion smoothing filter is developed, which aids in the low dy namic range image enhancement. The input image is decomposed into two layers using TV-L1 denoising method into base layer and detail layer and are sent into two channels, namely, structure pathway and detail pathway, which processes the low and high frequency information of the input visual images. In the structure-pathway, a normalization model incorporating the local and global visual adaptation terms is used, which manifested better performance in the case of the visual scenes with fluctuating illumination conditions. Furthermore, the detail enhancement and noise suppression are achieved in the detail-pathway based on local energy weighting. The out puts of two channels are unified and is eventually subjected to Perona Malik diffusion filter to achieve the low-light and night-time image enhancement. Experimental results on LDR image dataset called PKUnight corroborated that the proposed bioinspired image enhancement framework contributes well to the image enhancement tasks efficiently and outperform the related state of-the-art techniques in terms of standard image quality evaluation me