Adaptive Multi-Scale Retinex algorithm for contrast enhancement of real world scenes

Contrast enhancement is a classic image restoration technique that traditionally has been performed using forms of histogram equalization. While effective these techniques often introduce unrealistic tonal rendition in real-world scenes. This paper explores the use of Retinex theory to perform contrast enhancement of real-world scenes. We propose an improvement to the Multi-Scale Retinex algorithm which enhances its ability to perform dynamic range compression while not introducing halo artifacts and greying. The algorithm is well suited to be implemented on the GPU and by doing so real-time processing speeds are achieved.