直线特征保持的立体图像缩放

发布时间：2019-04-01 06:29

【摘要】：近年来随着3D显示设备的普及,内容感知的立体图像缩放问题得到了国内外学者的关注.与2D图像缩放不同,立体图像缩放不仅需要保持并突显图像显著区域,还必须保持立体图像视差的一致性.文中提出了一种立体直线特征保持的立体图像缩放的方法,通过在基于网格变形的图像缩放的方法中引入立体直线特征保持和内容感知的随机匹配点等约束条件,在保持立体图像视差的一致性的同时,较好地保持立体图像中直线区域的直线特征.首先,基于立体图像视差图,定义立体图像Hot-target图,检测立体图像显著区域,并获取立体图像直线特征;其次,引入内容感知的随机匹配点作为视差一致性约束;最后,构建基于四边域网格的受限能量模型,优化实现立体图像缩放.文中提供了若干实验数据说明该方法是有效的.
[Abstract]:In recent years, with the popularization of 3D display equipment, the problem of content-aware stereoscopic image scaling has been concerned by scholars at home and abroad. Different from 2D image scaling, stereo image scaling not only needs to maintain and highlight image salient areas, but also must maintain the consistency of stereo image parallax. In this paper, a method of stereo image scaling with solid line feature preservation is proposed. Constraint conditions such as stereo line feature preservation and content-aware random matching points are introduced in the method of image scaling based on mesh deformation. While maintaining the consistency of stereo image parallax, the straight line feature of straight line region in stereo image is well maintained. Firstly, based on the stereo image disparity map, the stereo image Hot-target map is defined to detect the prominent area of the stereo image and obtain the straight line features of the stereo image, secondly, the content-aware random matching point is introduced as the parallax consistency constraint. Finally, a constrained energy model based on quadrangle mesh is constructed to optimize the realization of stereo image scaling. In this paper, some experimental data are given to illustrate the effectiveness of the proposed method.
【作者单位】：广东外语外贸大学信息学院;国家数字家庭工程技术研究中心(中山大学);中山大学信息科学与技术学院;
【基金】：国家自然科学基金(61370160,61320106008,61402120) 广东省自然科学基金(2014A030310348) 广东省高等学校科技创新项目(2012KJCX0048) 广州市科技计划项目(2014J4100032)资助
【分类号】：TP391.41

【参考文献】