Underwater image enhancement by combining color constancy and dehazing based on depth estimation

Abstract

The physical properties which are present in the underwater environment affects the images captured by the visual sensors. As a consequence of these properties, the captured image includes non-uniform illumination. This non-uniform illumination cause color distortion, low contrast, white regions, and color casts. An underwater image enhancement method is proposed by combining a color constancy framework and dehazing. In the color constancy framework, both the white patch retinex and gamma correction are used to illuminate a non-hardware based balanced artificial illumination in the reference image. A chromatic adaptation technique (CAT) is adapted to correct the color cast caused by the non-uniform illumination. The color transferred image is then transformed into HSI with gamma correction in the Intensity (I)-component. This gamma correction enhances the intensity of the color transferred image. The gamma-corrected HSI image is converted to an RGB image. The dehazing is based on the estimation of artificial background light and transmission map depth. The depth is estimated from the difference of channel intensity prior (DCIP), which is the difference between the maximum and minimum intensity priors. Further, normalization of the DCIP with the histogram stretching enhances the contrast of the estimated depth. A saturation correction factor is proposed for color correction. This correction factor, estimates the artificial background light, and solves the non-uniform illumination limitations in the turbid image. A guided and rolling guidance filter is adapted to refine the estimated transmission map depth. Finally, the recovered image is transformed into HSI image with gamma correction on the I-component. The gamma-corrected HSI image is transformed to an RGB image. The recovered RGB image results with enhanced contrast and brightness. The proposed method enhances the contrast, preserves the visual information such as texture smoothing, edge-preserving, no halo effect, and decreases artifacts. We experimented the proposed method with that of the existing methods and observed that the proposed method resulted in a substantially improved image quality for the human visual perception.