An Improved Retinex Method for Low Light Image Enhancement

Enhancement of low-light image is difficult because it must account for not just brightness recovery but also more sophisticated concerns such as colour distortion and noise that are often hidden in dark. Simply increasing the brightness of a low light image can accentuate such artefacts. Image enhancement in low-light circumstance from the surveillance system plays a very important role for the security purpose. This is an active research topic, where many algorithms are proposed for magnifying the intensity of dark images. To overcome this issue a unique end-to-end attention-guided technique based on retinex is proposed. Here many effective image enhancement methods are ground on retinex theory. Subsequently diverse algorithms are used such as single scale retinex (SSR), multiscale retinex (MSR), multiscale retinex with color restoration (MSRCR) models.  color space model is applied with retinex algorithm. To enhance the quality and brightness of image; gamma correction with multiple values are used before applying Improved Multi Scale Retinex with CIELAB color (IMSRLab) on image. Extensive tests on standard LOL datasets show that our technique can deliver high fidelity enhancement outcomes for lowlight images, and that it outperforms existing state-of-the-art methods both statistically and visually.