2d color image enhancement based on conditional generative adversarial network and interpolation

Abstract

In the rapid development of autonomous driving technology. Precise detection of objects might assist self-driving cars to drive as safely as human. The object detection is frequently uses point clouds and produces high quality environment color images to match. However, at night or when the light is dim, it affects the quality of color images. In order to overcome this, the existing image enhancement is focused on the histogram equalization method [1] and Retinex algorithm [2]. This paper proposes to use the Conditional Generative Adversarial Network (cGAN) [3] to train the intrinsic images for quickly decomposed shadow layer, and then use the interpolation method to achieve the image contrast enhancement.