An overview of night vision colorization techniques using multispectral images: From color fusion to color mapping

Multispectral images present complimentary information, which enables night vision (NV). Specifically, night vision colorization using multispectral image increases the reliability of interpretation, and thus they are good for visual analysis (human vision). The purpose of NV colorization is to resemble a natural scene in colors, which differs from false coloring. This paper gives an overview of NV colorization techniques proposed in past decade. Two categories of coloring methods, color fusion and color mapping, are discussed and compared in this paper. Color fusion directly combines multispectral NV images into a color-version image by mixing pixel intensities. A channel-based color fusion method will be reviewed. Color mapping usually maps the color properties of a false-colored NV image (source) onto that of a true-color daylight picture (target). Four coloring mapping methods, statistical matching, histogram matching, joint histogram matching, and lookup table (LUT) will be presented and compared. The joint histogram matching is newly introduced in this paper. The experimental NV imagery includes visible (RGB), image intensified, near infrared, long wave infrared. From the experimental results, the following conclusions can be made: (i) The segmentation-based color mapping method produces the most impressive and realistic colors but it requires heavy computations; (ii) Color fusion and LUT-based methods run very fast but their results are less realistic; (iii) The statistical matching method always provides acceptable results (i.e., never fails); and (iv) Histogram matching and joint-histogram matching can generate more impressive colors when the color distributions between source and target are similar.