Human visual enhancement using Multi Scale Retinex

The Human Vision System (HVS) is responsible for color constancy and dynamic range. It is reasonably tolerant to variation under strong lighting conditions; however, this is not the case for camera images. We present a new six-step technique to enhance colored images that have been degraded due to non-uniform illuminations, using Multi Scale Retinex (MSR): (1) principal component analysis is employed on the colored (RGB) image to extract the desired luminance and chrominance, (2) the luminance is enhanced by the MSR scales, (3) the Discrete Fourier Transform (DFT) is introduced to optimize the enhancement process, (4) an enhanced luminance ratio is computed, (5) RGB values of image are computed in a ratio of the enhanced luminance with initial RGB values, and (6) the final enhanced image is attained using a contrast stretching approach. The proposed MSR image enhancement algorithm is a simple, automatic, and fast technique, and provides dynamic range compression with precise color rendition. Extensive simulations are performed on 994 colored images captured in outdoor environments. The proposed technique is shown to obtain the desired rendering in terms of fine details and naturalness.