HSV Fusion Of Near-infrared Image And Visible Image For Scene Recognition Via Sparse Recognition Using Intra-class Dictionary

As an important branch of pattern recognition,natural scene recognition has a very important role and value in related fields such as unmanned vehicles and robots.However,the scene image is complex,changeable and not rich enough,making the accuracy of scene recognition difficult to improve.The multi-source image fusion can improve the reliability and robustness of the actual application system,improve the utilization efficiency of image information,and use the redundant complementary information contained in different sensors can obtain a more comprehensive and accurate description and identification of the detection scene and target.With the popularization of the visible-near infrared common platform imaging system,intelligent natural scene recognition is becoming the current research hotspot.For this reason,aiming at the requirement of intelligent observation of typical natural scenes,this article proposed a sparse scene recognition method via intra-class dictionary that combines near-infrared image and visible image.First,the visible light image was transformed into the HSV space by the luminance image coding;then the luminance channel in the HSV space was replaced with the near infrared channel image;and finally the replaced image was restored to the RGB space,and a fused color image was obtained.Then,the global GIST features and local PHOG features were extracted and fused,which can be combined with sparse recognition using intra-class dictionary to obtain the scene classification results.The experimental results on the RGB-NIR database show that the recognition accuracy of the fused image is significantly higher than that of the unfused image,which proves that the fusion of near-infrared images does make the scene information more abundant.