Abstract: In order to improve the visual perception of low-light image, a kind of low-light image
enhancement algorithm based on iterative multi-scale guided filter Retinex was proposed. Firstly, the
low-light image was converted into YCbCr color space. Secondly, the illumination image of the Y
channel was estimated with iterative multi-scale guided filter. By increasing the scale parameter and
the smoothing parameter iteratively in exponential form, the edge was preserved and details of
different contrasts was eliminate. And then, the reflected image was obtained by the estimated
illumination image and the contrast correction procedure. Finally, the image was converted into RGB
color space, and this is the final enhanced image. Experimental results show that our method can better
obtain the illumination image so that the visual effect of the enhanced image can be improved obviously.

Key words: low-light image, Retinex, guided filter, edge-preserving and smoothing, multi-scale