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
