With the development of mobile phone cameras, more and more people use their mobile phones to take daily photos. However, due to the inherent limitations of the size and performance of the sensor and light of the scene, the imaging effect is often unsatisfactory. Therefore, we must make afterwards image enhancement on poorly performing images so that the images can be adjusted in terms of brightness and color. The application of deep learning in computer vision has developed rapidly in recent years, and together with the traditional rule-based approach, it has become two major approachs in the field of computer vision, and both have their own advantages and disadvantages.We propose learning-based image enhancement method combined with rule-based method, which can effectively enhance the brightness and color of images with different brightness.We first classify the original image according to the brightness, and use the adaptive gamma correction to automatically complete the brightness correction of the original image with too high or low brightness. Next, we use Cycle Generative Adversarial Networks trained by unpaired original image and enhanced image datasets to achieve the goal of color enhancement. Finally, we can get the enhanced image with appropriate brightness and vivid color. The experimental results show that the proposed method can effectively enhance the original images of various brightnesses. In addition, we also experiment and discuss the architecture of Cycle Generative Adversarial Networks, and finally design a more efficient Cycle Generative Adversarial Networks structure after evaluation. Experimental results demonstgrate the effectivess of the proposed method.
