为了获得一种具有艺术视觉效果的镶嵌图象,提出了一种基于多尺度小波分解的图象镶嵌技术,该技术首先对原始图象的各子块区域和图象库中的每一幅图象进行多尺度小波分解;然后逐层计算各图象小波分解系数的标准方差和它们之间的相似距离,并据此从图象库中选取与原始图象中的各子块区域最佳匹配的贴图,再将其镶嵌到原始图象中的对应区域;最后对贴图进行逐像素的颜色校正,使贴图的颜色与原始图象尽可能一致.同时,根据人眼观察某个区域时往往通过取整或将细节取平均来得到一个总体强度效果这一视觉特性,使最终的镶嵌图象具有在近处看到的是各个贴图的内容,而在远处观看则是原始图象的总体轮廓的视觉效果.利用计算机来自动地生成这种镶嵌图象的实验结果表明,该方法是有效的.

In this paper a novel image mosaicking approach is proposed, which includes three steps. Firstly multi-scale wavelet decompositions are performed on each image segment that is partitioned from large reference image and each tile image in image library respectively. Then standard deviations of image wavelet decomposition coefficients and similarity distance between image segments and tile images are calculated in each level, which latterly are used as a criterion for tile image selection that images in library best matched to the mosaic area in reference image are selected. Finally to make the color of mosaic image identical to that of reference image, color correction on each pixel of mosaic image is employed. Due to human visual character that what human perceives is only the whole intensity of observed region by rounding and meaning of the intensity values in it, the final work of mosaicing has realized the effect that the tile images only can be discern nearly and the overall figure looks similar to that of the reference image from a distant view. An automatic computer mosaic image generation routine using above algorithm has been implemented, which shows the feasibility of this approach.