Abstract: To deal with the problems of low definition and unnatural color after infrared image colorization, this paper introduces an improved local linear embedding algorithm into the application of infrared image colorization, and proposes an infrared image colorization that features local linear embedding and fuzzy c-means clustering. Firstly, by enlarging the neighborhood range and adding weight information, the algorithm improves the defect of the local linear embedding algorithm which is sensitive to sparse matrix. In the space characterized by pixel vectorization of infrared and color template, Nearest Neighbor Search is used to figure out the best match coefficient. The color of color template can be mapped onto the specific areas on infrared image by computing color value, then color transfer from the template to the targeted infrared image done. The improved fizzy C-average clustering is carried out to conduct color clustering on infrared image, and on the color clustering set, histogram equalization to equalize segmented color, after which equalized image is finally synthesized. After the simulation comparison between this algorithm and the other two infrared colorization algorithms, the result shows that the proposed infrared image colorization algorithm can achieve clearer and target-prominent colorization only by target infrared images and color templates.

Key words:  infrared image, colorization, LLE, fuzzy clustering C-mean, peak density, histogram equalization