基于多分辨率色彩传递的彩色夜视方法研究

Color Night Vision Research Based on Multi-resolution Color Transfer

为了获得符合人眼感知的彩色夜视图像,帮助观察者获取丰富的场景信息和舒适的观察效果,提出了一种基于多分辨率色彩传递的彩色夜视处理方法.结合微光与红外图像特点,对二者进行YUV空间拮抗视觉特性的彩色融合之后,采用控向金字塔对彩色融合图像(源图像)和参考图像各颜色通道进行多分辨率分解,计算子带图像的均值和标准方差,通过参考图像与源图像在各子带的方差比调整源图像的子带系数值,再经金字塔重构,最终使微光与红外彩色夜视图像获得类似参考图像的自然感色彩.分析了lαβ、YUV和RGB颜色空间对色彩传递的影响,并确定YUV颜色空间更适于多分辨率分解的色彩传递.相比传统的线性色彩传递方法,本方法不仅使彩色夜视图像拥有更符合真实场景的色彩,还可以提高其细节信息,改善场景感知,提升融合图像目标探测能力.

In order to acquire color night vision images accordant with human perception,and help observers to get more abundant scene information,a kind of color night vision method based on color transfer in multi-decomposition structure is proposed.Concerning their characteristics,low-level light visible and IR images are combined into a color fused image （source image） based on opponent vision property in YUV color space.Then the steerable pyramid is used to decompose the fused image and a colorful day-time reference image into multi-scale sub-band images.The mean and standard deviation of each sub-band image are calculated.Each sub-band coefficient of source image is modified according to the deviation ratio of the reference image to the source image in respective color channels.Finally,reconstruct the transferred image and obtain a low-level light visible/IR color night vision image with comfortable color appearance similar to the reference image.The transferred results in lαβ,YUV and RGB space are analyzed.YUV space is recommended to be the best color space for multi-decomposition color transfer.Comparing with traditional linear color transfer,the proposed method can make a color night vision image more realistic and natural,enhance the details,improve the situation perception and target detection ability.