In this work we propose a network model for the retinex computation. This model provides a better insight on the retinex theory. By drawing analogy from neuro-physiological activities in human vision system, we have suggested to replace the logarithmic function used in earlier models by a sigmoidal function. It has produced good color rendition. In addition in this case we do not require any color restoration filtering as a post-processing stage. Considering the interactions between rod and cone cells through the bipolar cells in the human vision system, we have further proposed the retinex computation using the diffusion of luminance component only. We found that this has produced almost similar color constancy and color rendition in the processed images.