基于伪彩色融合图像聚类的夜视图像上色算法

Colorizing Algorithm of Night-Vision Image Based on Clustering of False Color Fused Image

提出一种将小波融合和基于伪彩色融合图像的C均值聚类用于图像色彩传递算法中的彩色夜视方法。在色彩传递前将可见光和红外图像进行小波融合得到灰度融合图像作为目标图像,保持了较好的纹理信息和目标信息;再对彩色源图像进行基于连接相对熵的彩色阈值分割;然后针对灰度融合图像的特点,根据一种基于伪彩色融合图像的C均值聚类方法,将伪彩色融合图像的彩色信息作为特征向量应用在夜视图像的分类当中,得到较好的分类效果,并基于此分类结果再进行色彩传递,得到更为自然的彩色夜视图像。实现了对夜视图像的自动色彩传递,得到的图像色彩较真实,纹理清晰,将有利于人眼的目标识别。

Using wavelet fusion and C-means clustering of false color fusion image, a color night-vision method for color transfer algorithm of image is presented. Firstly, before transferring color to the night vision image, the fusion of visible image and infrared image based on wavelet transformation leads to the gray fused image as target image, which keeps texture information and target information well. Secondly, the source color image using color threshold segmentation based on joint relative entropy is segmented. Finally, a C-means clustering algorithm based on the false color fused image is proposed , according to the characteristics of the gray fused image. In this algorithm, the color information of the false color fused image is used for the night vision image clustering as feature data vectors, and the segmentation result is good. Then, transferring color to the night vision image based on the clustering results is carried out and natural colour night vision image is obtained. The automatic color transfer of the night vision images is realized. The color appearance of the obtained image is natural and its texture is also clear. These results indicate that the proposed method will benefit the target identification for human eyes.