利用图像块数据库和纹理特征点的图像彩色化方法

Image Colorization with Block Image Database and Texture Feature Points

针对颜色转移彩色化算法存在速度慢、效果不佳及人工干预性强等问题,提出一种新型的彩色化算法.首先从聚类分割后的各类区域彩色图像中提取矩形块,并计算图像块的颜色直方图和纹理特征;利用颜色直方图比较图像块的相似性,实现数据库的选择性录入;利用目标图像块与彩色图像块纹理特征点之间的欧氏距离比较,查找最佳匹配的图像块,进而实现目标图像的彩色化.通过建立树木、天空、沙滩和草地4类图像块数据库,实现基于数据库技术的免除人工干扰的图像彩色化.

We propose a new method for image colorization to overcome problems including heavy computation load,poor quality,and excessive human intervention.We choose rectangular blocks from various color images using segmentation and clustering,and calculate color histograms and texture feature points of the blocks.We then build a database by comparing similarity of the blocks based on the color histograms,and find the best matching in terms of the Euclidean distance between the target gray-scale blocks and the source color blocks.Having constructed databases of four kinds of image blocks,tree,sky,sand and grass,the target gray image can be colorized using the block database without human intervention.