Abstract: According to the multi-scale Retinex algorithm (MSR) defect images peak at low SNR, this paper puts forward a method of noise suppression in low luminance region according to the theory of Retinex, the characteristics of illumination and reflection components, the two respectively by different nonlinear functions to adjust, to enhance the image contrast and enhance image detail. The experimental results show that, in HSV and RGB color space, after using this algorithm in low light, uneven illumination image case, the peak signal to noise ratio and the image contrast are higher than that of MSR algorithm, and the processing speed is faster than MSR algorithm.

Key words: Retinex, peak signal-to-noise ratio, noise suppression, color space