基于场景识别的夜视图像彩色融合方法

Night vision image color fusion method based on scene recognition

将红外与微光图像进行彩色融合可以使目标检测和识别更精确。针对现有的使用固定参考图像的彩色传递融合方法,提出了一种新的结合场景分类、融合质量评价以及彩色传递的融合方法。首先提取输入图像的GIST特征并利用SVM分类器进行场景的分类,随后利用彩色融合质量评价方法在相应类别的图像库中匹配最佳的参考图像,最终使用彩色传递的方法将红外与微光图像融合为彩色图像。仿真结果表明,相较于其他彩色融合方法,该方法在无须观测环境先验信息的情况下,能够在线性以及非线性彩色空间使融合图像接近自然真实的色彩感觉,更易于分辨识别目标,从而达到提高机器视觉效率的目的。

Infrared and low light level images color fusion can make the target detection and recognition more accurate.Aimed at the existing color fusion methods based on color transfer,this paper proposed a new fusion method that combined scene classification,fusion quality evaluation and color transfer.Firstly,it extracted GIST features of the input image and used SVM classifier to do the scene classification,then it used the color fusion quality evaluation method to find the best matched reference image in the image library of the corresponding category,finally it got the color fusion result of infrared and low light level image using color transfer method.The simulation results show that compared with other color fusion methods,this method doesn’t need prior information of observation environment,and can display the fusion result in natural colors under both linear and non-linear color space,which makes the target identification more accurate,therefore improving the efficiency of the machine vision system.