Near infrared image colorization

• Main Author: You, Xinchi
• Other Authors: Lap-Pui Chau
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: https://hdl.handle.net/10356/141306
• Institution: Nanyang Technological University

Near-infrared images are used in low-light conditions and night vision environments, playing an important role in the fields of traffic driving, security monitoring, and military exploration. It reflects the radiation information of the target scene. Compared with visible light images, the information is seriously missed, especially for the color information. Improving the quality of the collected images requires huge costs. According to the biological vision mechanism, humans can distinguish more than twenty different gray levels from black to white, and cone cells can distinguish thousands of different levels of color, so the sensitivity of the human eye to grayscale images is far less than that of color images. In addition, the infrared image is more blurred than the grayscale image, so the image cannot directly convey the real situation. The prediction of daytime natural light scenes from nighttime infrared images is a very popular image translation problem in deep learning. This paper mainly uses the Generative Adversarial Networks model to solve the problem of generating nighttime infrared images into daytime color images. This paper first analyzes the characteristics of traditional infrared night vision colorization, and then finds that there are few methods involving night vision infrared daylight colorization, and the network generates better results to solve similar problems. Then, we introduce some traditional methods in colorization. However, in recent years, neural network are widely used in this field especially Generative Adversarial Networks. In my own part, UNET is regarded as the generative network and net D is just convolutional layers. What’s more, to preserve the edge and texture information of the infrared images, edge extraction method is used for image preparing.