融合多线索信息的数字图像抠图方法研究  

Research on digital image matting method based on fused multiclue information

作　　者：张超[1] 都玉莹 韩成[1] 白烨[2] ZHANG Chao;DU Yuying;HAN Cheng;BAI Ye(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China;College of Optical and Electronical Information,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022 [2]长春理工大学光电信息学院,长春130022

出　　处：《计算机工程与应用》2018年第17期169-174,185,共7页Computer Engineering and Applications

基　　金：吉林省科技攻关计划项目(No.20170203003GX;No.20170203004GX)

摘　　要：针对自然图像抠图方法中存在对先验知识过度依赖和交互输入繁琐的问题,为了扩展自然图像抠图方法的使用范围,提升自然图像抠图方法的自动化程度,提出一种融合多线索信息的数字图像抠图方法。利用原始自然图像所对应的深度信息和视觉显著度信息进行感兴趣区域粗分割;利用形态学的膨胀与腐蚀算法对感兴趣区域的分割结果进行粗分割区域膨胀和粗分割区域腐蚀操作,从而得到抠图过程所需的三分元素图;利用彩色纹理图像和三分元素图,并结合使用相似性传递抠图方法获得精细的前景目标抠图结果。实验结果表明,该方法不仅能够得到较为理想的抠图效果,而且大大提升了自然图像抠图方法的自动化程度。In the natural image matting method, there is a problem that the prior knowledge is excessively dependent and the interactive input is complicated. In order to extend the use of natural image matting methods and improve the automation of natural image matting methods, a digital image matting method based on multi-clue information is proposed. First of all, it uses the depth information and visual saliency information of the original natural image to segment the interest region roughly. Then, the morphological expansion and erosion algorithms are used to expand and erase the rough segmentation result, and it can get the trisectional element map of the matting process. Finally, the fine foreground object matting result is obtained by using the color texture image and the trisectional element map together with the similarity transfer matting method. The experimental results show that this method can not only achieve the ideal matting effect, but also greatly enhance the automation of the natural image matting method.

关 键 词：数字化抠图 多线索 泊松抠图 三分元素图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]