A Low-illumination Image Enhancement Algorithm Based on Morphological-Retinex(MR) Operator

Low-illumination images were often caused by night environments, which are lack contrast and sharpness. Night image recognition, capture, or autonomous-driving systems require meritorious algorithms to enhance the low-illumination images. The Retinex-based method has been used in image enhancement universally, but previous literature had some enhancement limitations. Since the earlier methods show poor robustness, the night recognition and low-illumination image training programs were prone to fail. For solving those problems, a new algorithm called Morphology-Retinex (MR) is proposed to enhance the details and edges of the low illumination images. We also offer a new automatic color restoration method to restore the authentic color of the object. As a result, the scores of PNSR and SSIM are 15% higher than traditional methods, and the clarity is excellent, which is about 50% higher than similar algorithms in recent years. This algorithm can improve the quality of low-illuminance image data sets and reduce the difficulty of image recognition. When training machine learning models using low-illumination image data sets, this method will increase the success rate.