摘要
结合图割算法,提出了一种针对低景深(Depth of field,DOF)图像的自动分割模型.首先,通过改进的点锐度算法得到图像的点锐度图,并结合图像的颜色特征,得到一个四维的特征向量.其次,通过对图像点锐度图强边缘的计算,利用图像清晰部分边缘较连续,模糊部分边缘较弱、连续性较差的特点得到图像初步的前景/背景区域.然后,对前景/背景的颜色和点锐度特征进行高斯混合模型(Gaussian mixture model,GMM)建模,结合全局、局部自适应的λ值,对图割算法的Shrinking bias现象进行改善.最后,通过迭代的图割算法对前景/背景区域进行修正.实验结果表明,该模型鲁棒性较高,分割结果更加精确.
An automatic segmentation model combined with graph cuts algorithm for low depth of field （DOF） images is proposed. Firstly, the point sharpness algorithm is improved to extract the point sharpness map of the image. In combination with color features, a four dimensional vector is constructed. Secondly, strong edges of the point sharpness map are exacted and the characteristics that the edges of clear part of an image are commonly continuous and the edges of blurred part are weak and discontinuous are used to get the preliminary foreground/background regions. Then, Gaussian mixture model （GMM） is used to model the features of point sharpness and color and by using global and local adaptive the shrinking bias problem of graph cuts algorithm is improved effectively. Finally, the iterative graph cuts algorithm is used to revise the foreground/background regions. Experiments show that the proposed segmentation model is more robust and more accurate.
出处
《自动化学报》
EI
CSCD
北大核心
2015年第8期1471-1481,共11页
Acta Automatica Sinica
基金
国家自然科学基金(61473157)
江苏省高校自然科学研究项目(13KJ B520013
14KJB520019)资助~~
关键词
图割
低景深
点锐度图
高斯混合模型
Craph cuts, low depth of field, point sharpness map, Gaussian mixture model （GMM）
