DSP Implementation of Contrast Enhancement by Nonlinear Diffusion

Abstract

A real-time image enhancement algorithm is usually needed by an image processing system to improve the visual quality of the captured image sequence. Contrast enhancement by nonlinear diffusion (CEND) in the literature shows excellent performance in artifact removal and local contrast enhancement. However, CEND algorithm is too complex to be implemented on real-time systems. In this paper, in order to achieve real-time image enhancement algorithm based on CEND, we choose C674x DSP (Digital Signal Processing) as targeted device and perform code simplification as well as in-depth code optimization, including compiler-based optimization, calculation simplification and on-chip memory utilization. These optimization measures do not involve access to peripherals, which makes the algorithm stable and convenient to be transplanted to other DSP platforms. We test the program of the CEND algorithm on TMS320DM8168 evaluation board and have verified that the implemented algorithm as a single thread on DSP subsystem is able to achieve above 25fps for 640×480 sized image sequence.