摘要
针对经典贝叶斯抠图方法在函数建模和模型求解方面存在不足,提出基于纹理分布弱假设和正则化策略的抠图方法:首先针对函数建模问题,基于直方图和巴氏距离,定义和设计了对高斯分布方差进行修正的计算测度以及修正系数,提出一种考虑纹理复杂程度的自适应方差高斯分布模型;然后针对模型求解方法的不足,提出一种基于正则化策略的抠图求解模型,即通过增广拉格朗日乘子法,在基本模型中增加了数据约束项与惩罚项.通过定性和定量相结合的方式进行实验,结果表明该方法在复杂纹理区域上取得更好的效果,并且更加接近真实值.
The Bayesian matting method has two drawbacks in function modeling and solving method. First, we design a computation measure to correct the fixed variance and propose a Gaussian model with adaptive variance to reflect the complex distribution of nature images texture. Second, we proposed a solving method based on the regularization strategy, which adds data constrain term and penalty term into the original method through the Augmented Lagrangian Multiplier. Finally, the experimental results show that our method outperforms other methods in complex texture areas.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2016年第10期1672-1678,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61472289)
湖北省自然科学基金(2015CFB254)
关键词
自然图像抠图
纹理分布
约束优化
正则化策略
image matting
texture distribution
constraint optimization
regularization strategy
