文献类型：期刊文献

中文题名：基于类别色彩查找表的彩色夜视方法

英文题名：Night-Vision Colorization Based on Category-Oriented Color Lookup Tables

作者：张玮雯[1];谷小婧[1];顾幸生[1]

机构：[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室

年份：2019

卷号：45

期号：6

起止页码：954

中文期刊名：华东理工大学学报（自然科学版）

外文期刊名：Journal of East China University of Science and Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金（61775058,61573144,61773165）

语种：中文

中文关键词：彩色夜视;类别色彩查找表;语义分割;多波段图像;热红外图像

外文关键词：color night vision;category-oriented lookup tables;semantic segmentation;multispectral image;infrared thermal image

摘要：针对夜视图像解读困难的问题,提出了一种面向类别的色彩查找表的彩色夜视方法。该方法可以将不同景物的特征色彩映射到多波段夜间图像,使观察者能够更快、更好地解读图像,从而提高态势感知并缩短反应时间。这种映射方法首先采用深度学习中的语义分割技术对图像进行分割,然后使用类别色彩查找表固定不同波段对应色彩,避免了全局彩色化时出现的色彩单一、不自然的缺点。实验结果表明,与其他全局彩色化方法相比,本文方法对于夜视图像的解析能力更强。
Due to the low illumination in the night, it is difficult to observe and identify the nighttime images exactly. To address this problem, we try to map the feature colors of different scenery to the nighttime images of visible and long-wave infrared bands such that the observers can analyze the images and speed up the reaction time. To this end, this paper presents a category-oriented color lookup tables night-vision colorization method by combining deep learning. Firstly, it utilizes the semantic segmentation via parallel fully convolutional network and the feature of twoway network at the end of the network to get the segmentation categories. And then, it colorizes the nighttime images by each category-oriented color lookup table to avoid the same color scheme and unnatural colors resulted from the global colorization like using single lookup table. In the post-processing stage, both the image normalization and the multispectral bilateral filtering will be adopted to smooth the colors and preserve the edges. The proposed method can fix the dual-band information, which is different from the other methods. By constructing the category-oriented color look-up table, the proposed method can obtain more stable results and much better material resolution. Moreover, the image contrast is enhanced and the thermal targets is highlighted by red such that it is much easier to detect. Besides,the proposed method can also be extended according to different environments and applications via the fixed categoryoriented color lookup tables. Finally, by the comparison with other methods, such as global colorization by single lookup table, it is shown that the proposed method in this work can attain significant improvements in this area of nightvision colorization.