Local color correction of stereo pairs

The interest in the production of stereoscopic contents is growing rapidly. Stereo material can be produced using different solutions, from high level devices to standard digital cameras suitably coupled. In the latter case, color correction in stereoscopic images is complex, due to possible different Color Filter Arrays or settings in the two acquisition devices: users must often tune each camera separately, and this can lead to visible color inter-differences in the stereo pair. The color correction methods often considered in the post-processing stage of stereoscopic production are mainly based on global transformations between the two views, but this approach can not completely recover relevant limits in the gamuts of each image due to color distortions. In this paper we evaluate the application of perceptually-based spatial color computational models, based or inspired by Retinex theory, to pre-filter the stereo pairs. Spatial color algorithms apply an unsupervised local color correction to each pixel, based on a simulation of color perception mechanisms, and were proven to effectively reduce color dominants and adjust local contrasts in images. We filtered different stereoscopic streams with visible color differences between right and left frames, using a GPU version of the Random Spray Retinex (RSR) algorithm, that applies in few seconds an unsupervised color correction, and the Automatic Color Equalization (ACE) algorithm, that considers both White Patch and Gray World equalization mechanisms. We analyse the effect of the computational models both by visual assessment and by considering the changes in the image gamuts before and after the filtering