Research On Algorithm Of Low-light-level Visible And Thermal Infrared Imge Fusion Based On Color Transfer And Multi-scale Decomposition

The low-light-level visible imaging and thermal infrared imaging,as the two main means of the current night vision technology,are key technologies developed by countries around the world.It is difficult to consider both scene detail performance and target detection performance by the low-light-level imaging or thermal infrared imaging system alone.In addition,the thermal infrared image and the visible image have good information complementarity of target scenes.Therefore,visible(low-light-level)and infrared image fusion technology,as the forefront of the current development of disciplines,is one of important research contents of multi-band image fusion technology and has been successfully applied in the field of night vision imaging.In order to fully exploit the advantages of low-light-level visible and infrared image fusion,the detail(edge)information of the source image is further preserved and the target features are highlighted.Thus,the night-vision image filtering and enhancing method,image fusion method based on color transfer,fusion method based on multi-scale decomposition and target detection/recognition capabilities for the fusion are further studied in this thesis to improve the visual perception.The main contents of the research include:(1)A noise-intensity estimate method based on statistical characterisitics is proposed for low-light-level video images.Then,based on the gradient domain guided filtering(GDGF),an adaptively denoising algorithm for low-light-level video images(GDGF-AD)is proposed in this study.The noise reduction effect of the GDGF-AD algorithm has certain advantages over some of the current advanced methods.The GDGF-AD algorithm has a relatively small amount of computation and a clear advantage in processing speed,and it has strong real-time potential,which is of great significance to the practical application of night-vision imaging.Meanwhile,the application of GDGF to detail enhancement processing of infrared images is studied.Compared to guided filtering,GDGF-based image detail enhancement can achieve better performance.(2)The effectiveness and feasibility of the color image fusion algorithm based on color transfer as a grayscale image fusion algorithm are studied.Experiments show that grayscale image fusion is a feasible way through YUV color space.The grayscale image fusion based on YUV color transfer has obvious advantages in contrast and sharpness compared to some classical grayscale fusion algorithms.The proposed algorithm has been realized in real-time on the image processing platform,which integrates the natural-appearance color fusion and grayscale image fusion processing.Thus,appropriate fusion mode can be selected according to the observation needs,which provides a flexible choice for high-performance color/grayscale image fusion system.(3)Based on the idea of color image fusion with color transfer,a luminance stretching color transfer(LSCT)algorithm for the low-light-level visible image is proposed.LSCT algorithm can effectively improve the contrast of night vision images and obtain natural-appearance color.Experiments show that the LSCT algorithm can obtain better visual perception for human eyes,and the operation amount is small.LSCT algorithm has been implemented on the hardware platform in real time and has a wide range of application prospects.(4)The image decomposition structure with hybrid multi-scale based on GDGF is firstly proposed.Then,the algorithm of hybrid multi-scale decomposition based on GDGF(HMSD-GDGF)for grayscale fusion is proposed.Experiments show that the HMSD-GDGF algorithm has certain advantages in terms of fidelity of salient information,preservation of edge information,and the visual perception of the human eyes.In addition,the HMSD-GDGF algorithm can automatically select the best decomposition scale factor according to the size of the source image,which has relatively strong universality.Meanwhile,the night-vision context-enhancement(FNCE)algorithm based on the HMSD-GDGF algorithm is further proposed.The visual effect of night-vision source images are adaptively enhanced.Finally,the HMSD-GDGF algorithm is introduced to the fusion processing and the parameters of the fusion are improved considering the features of night-vision applications.The FNCE algorithm can achieve relatively better night-vision enhancement.(5)A portable low-light/long-wave infrared image fusion experiment system is developed.Through subjective evaluation experiments of target detection/recognition,the relevant experimental data are analyzed from three aspects: target detection/recognition probability,the influence of different observation distances,and the influence of different scenes.The following conclusions can be drawn: compared with single low-light-level or single long wave infrared channel,proper fusion methods will have obviously enhancement effects on target detection/recognition;and different color performances of color fusion will also affect target detection/recognition capabilities;although inappropriate color fusion may reduce the target detection/recognition capability,there is a color fusion mode that has a significant enhancement effect on most scenes.It has a certain guiding significance on how to further improve target detection/recognition capabilities.Supported by the National Natural Science Fund and other projects,relevant image fusion technologies of low-light-level visible/infrared are studied in the thesis,which provides theoretical and key technical support for the further development of color night vision.Some of the proposed algorithms have been successfully applied in some latest weapons and achieved good results.