Stereo correspondence for underwater images using adaptive weight-based cross-correlation

In this paper, we introduce an approach called as adaptive weight-based cross-correlation method to estimate the disparity map for a pair of underwater images. A majority of the existing stereo correspondence algorithms assume that the corresponding color values are similar to each other. However, it is not in the case of underwater images; variations in corresponding color values are very high due to factors such as, attenuation and scattering of light in underwater environment. The attenuation and scattering of light cause variations in the measured color of each scene point at its projections on the two images, which leads to radiometric variations between a pair of images. We employ comprehensive color image normalization method to normalize the input color images. The conventional correlation methods compute the disparity at a given point depends only on intensity values within a local window, but these techniques are sensitive to illuminative variations and relatively lower performance in the featureless regions. Therefore, instead of intensity value, we use color dependent adaptive weights computed for each pixel lies within support window, and the cross-correlation measure between the support window is employed to estimate the disparity. The experiments are evaluated qualitatively, and the result reveals that our method outperforms the conventional correlation methods under conditions such as variations in illumination and blur.