结合区域特性的图像改色算法  

Recolorization employing region cues

作　　者：张可为[1,2] 申群太[2] 

机构地区：[1]湖南工程学院电气信息学院,湖南湘潭411101 [2]中南大学信息科学与工程学院,长沙410008

出　　处：《计算机应用研究》2012年第6期2390-2392,共3页Application Research of Computers

基　　金：湖南省教育厅科研项目(08c220)

摘　　要：针对现有图像改色方法应用于零散分布对象时效果不佳的问题,提出了一种结合区域特性的两级协同扩散算法。首先将图像像素与通过过度分割得到的小块区域分别组成有权图的两级,除了建立像素级近邻连接外,图中还引入了区域级全连接,以及每块区域与其内部像素之间的连接;然后设计一迭代过程对所有这些连接进行信息调配,从而既能实现远距离扩散,也能使区域内像素的色彩特性趋近于一致。实验结果表明,即使添加非常少的人工标记该算法也可以获得很好的改色效果。Existing recolorization methods could hardly provide pleasant visual result when they were applied to scattered objects,this paper put forward a two-layers propagation algorithm incorporated with regions cues.Firstly,combined regions by over-segmentation with pixels to construct a two layer weight graph.In contrast to previous works which only employed connection between neighboring pixels,it introduced both full-connections in region layer and inter-connections between the pixels and their corresponding regions into this graph additionally.Then it integrated all these connections in a chrominance iteration formulation,not only facilitated propagation of local grouping cues across larger image areas but partly enforced color edit consistency inside each region.Experiment result indicates that even very small amount of user＇s scribble is good enough for this proposed method to produce wonderful visual effect.

关 键 词：改色 零散对象 有权图 协同扩散 

分 类 号：TP393.04[自动化与计算机技术—计算机应用技术]