Low-light video image enhancement based on multiscale Retinex-like algorithm

Abstract

The images taken in low-light environment always lose some details. Given that traditional multi-scale Retinex (MSR) algorithm always appears halo artifact in the edge area of the image, a multiscale Retinex-like is put forward for low-light video image enhancement. The proposed algorithm is operated in HSI color space. For Intensity layer, three Gaussian filters in traditional MSR algorithm are replaced by three guided filters to extract illumination component, then gets the reflection component, the sum of this two components as the new Intensity layer. For the Saturation layer, Gamma correction function is used to enhance Saturation, then the new HSI image is converted to RGB image. Four kinds of numerical methods are adopted to evaluate the images enhanced by different methods. The experimental results show that the proposed algorithm in terms of low-light image, not only makes the picture becomes clearer, but also extracts more image details. Furthermore the algorithm is more efficient.