针对现有基于全局统计信息的彩色夜视算法不能较好保持图像目标色彩恒常性的不足,提出一种简单有效的局部自然彩色夜视算法。首先依据人类视觉系统的中心-环绕对抗结构生成伪彩色融合图像; 按照不同颜色特征对伪彩色融合图像进行聚类,分割成不同颜色域,参考自然彩色图像中对应域为与融合图像对应位置像素集合;最后在简化Lab空间实现对应域间的局部颜色映射,将参考图像颜色特征传输至融合图像。实验结果表明,本文方法使得夜视图像获得如白天参考图像般自然、真实的效果,有利于提高场景辨识能力;方法执行速度快,便于实时处理。

In this paper we put forward a simple and efficient color mapping method to improve the sufficiency of nearly color night vision schemes based on global statistics.First,multi-band images are fused into the three components of an RGB image in terms of a simple linear fusion strategy based on the opponent property of ON-center OFF-surround reception in biological vision.Then a color clustering method is applied to produce color fields in the fused image.The corresponding fields in the reference image are obtained by locating the pixels in the fused image and finding the corresponding pixels in the reference image.Finally,the local mapping in the simplified Lab color space optimizes the match between the night vision and the reference images.Experiments have shown that the proposed scheme can yield a night vision image with a natural daytime color appearance while also improving the speed.