Research And Implementaion Of Reference-based Infrared Colorization Alogrithm

Infrared cameras are widely used in various scenarios such as home monitoring because the infrared light used in the imaging cannot be detected by the naked eye,so infrared cameras have strong concealment.However,due to the different imaging principles,infrared cameras can only generate black and white images,and the brightness of the objects in the infrared images is also different from the normal black and white photos.This feature affects the perception of infrared images and further hinders their direct use in other computational vision tasks.Therefore,colorization of infrared images to generate color images is of great value.In order to improve the quality of the near infrared image coloring results,we propose a reference based infrared coloring algorithm.The algorithm uses a color image related to the semantic content of the infrared image as a reference map to guide the algorithm to colorize the infrared image in order to generate the most reasonable result possible.Since the algorithm model cannot obtain the semantic connection between the infrared image and the reference image with full accuracy,it may lead to the generation of rough coloring results with wrong colors in some positions,and it is difficult for the coloring model to determine whether the colors are correct,so these wrong colors will further prevent the algorithm from generating reasonable color results.The algorithm proposed in this paper uses super pixels to reconstruct the rough coloring results generated by the algorithm model,so that the algorithm can finally generate reasonable color and edge-accurate results.Meanwhile,some low-exposure infrared images may generate some results with blurred details and with artifacts.To address this problem,a infrared coloring algorithm based on differentiable filter is proposed.The algorithm adds an image adaptive processing module to the previous algorithm,which uses differentiable filters to perform adaptive processing on infrared images to improve the effect of the algorithm model.We designed and implemented comparative and ablation experiments to verify the effectiveness of the above proposed algorithm.Qualitative and quantitative results show that the proposed algorithm has better results compared with existing grayscale coloring and image translation algorithms.