Night Vision Image Processing Based On Texture Transfer

Colorization of night vision images is a very important area of night vision research. However, the traditional technology of colorization only increased color information, do not solving the problems that night vision image is with low resolution and blurred details. Texture can simulate the real details of the scene surface features, and enhance realism drawn. Therefore, this paper proposed several night vision enhancement algorithms based on texture transfer and generation theory, the result images are with natural colors and textures.Night vision images including infrared images and low light level images, both have their own characteristics but also the shortcomings. Firstly, the characteristics of infrared images is less texture and contour information is better, low light level image of the rich texture and outline the characteristics of lack of information, extraction of the same scene infrared and low light level image texture features, the Laplace Multi-scale feature level fusion, are more prominent targets more clearly the texture of gray fused image; and fused images for visual assessment and quantitative indicators of qualitative evaluation. Evaluation showed that the fused image retains the respective infrared and low light level image advantages, objectives become more prominent and more detailed textures.Secondly, this paper proposed a natural color night vision approach based on texture synthesis. By transferring both texture and color information of the color image to the night vision image at the same time, we can acquire the better color night vision images. Sub-block from human interaction determined for the source image and target the natural color infrared image, the results of the texture generation kid block extended to the entire target texture image. And at last experimental result verify the feasibility and effectiveness of the algorithm.Finally, we extract different patterns of textures to construct a LLL image library including sky, road, grass and tree. Then we extract the texture features of each image in the LLL image library based on combination of GLCM and Gabor filter and construct the corresponding texture features library. And then, we classify each pixel of the original target LLL image through comparing the similarity of the pixel texture features with the texture features in the texture features library. We transfer colors and textures to the LLL image using image overprint method. The experimental results show that method based on texture feature library can be achieved more accurate low light level image segmentation and method based on the color texture fusion and overlay, in part to improve the lack of distortion of the scene depth of feeling issue that due to lost depth information in overlay. Experiment shows that the color LLL images with natural colors and textures and express more harmonious visual perception.