基于邻域扩展的半自动2D转3D方法
发布时间：2018-09-18 10:54
【摘要】：半自动2D转3D是解决当前3D影视内容匮乏的重要途径。现有方法大多借助局部邻域进行深度插值,忽略了图像的全局约束关系,因而难以准确恢复深度图的对象边界。针对该问题,提出邻域扩展的最优化深度插值方法。首先引入邻域的邻域,建立邻域扩展的最优化深度插值能量模型;其次在相似的像素点与其邻域加权深度平均值的差异近似相等的假设条件下,将深度插值能量模型的最优化问题转换成一个稀疏线性方程组的求解问题。实验结果表明,与当前流行的半自动2D转3D方法相比,该方法估计的深度图PSNR更高,同时增强了深度图的对象边界质量。
[Abstract]:Semi-automatic 2D to 3D is an important way to solve the shortage of 3D video content. Most of the existing methods use local neighborhood to carry out depth interpolation, ignoring the global constraint of the image, so it is difficult to accurately restore the object boundary of the depth map. To solve this problem, an optimal depth interpolation method based on neighborhood extension is proposed. Firstly, the neighborhood of the neighborhood is introduced to establish the optimal depth interpolation energy model of the neighborhood extension; secondly, under the assumption that the difference between the similar pixel points and the weighted depth mean of the neighborhood is approximately equal to that of the average value of the weighted depth of the neighborhood, the optimal depth interpolation energy model is established. The optimization problem of depth interpolation energy model is transformed into a sparse linear equation system. The experimental results show that compared with the popular semi-automatic 2D / 3D method, the proposed method can estimate the depth map with a higher PSNR and enhance the object boundary quality of the depth map.
【作者单位】： 南京理工大学泰州科技学院计算机科学与技术系;宁波工程学院电子与信息工程学院;
【基金】：国家自然科学基金资助项目(61170200) 浙江省自然科学基金资助项目(LY16F010014)
【分类号】：TP391.41
，
本文编号：2247684
[Abstract]:Semi-automatic 2D to 3D is an important way to solve the shortage of 3D video content. Most of the existing methods use local neighborhood to carry out depth interpolation, ignoring the global constraint of the image, so it is difficult to accurately restore the object boundary of the depth map. To solve this problem, an optimal depth interpolation method based on neighborhood extension is proposed. Firstly, the neighborhood of the neighborhood is introduced to establish the optimal depth interpolation energy model of the neighborhood extension; secondly, under the assumption that the difference between the similar pixel points and the weighted depth mean of the neighborhood is approximately equal to that of the average value of the weighted depth of the neighborhood, the optimal depth interpolation energy model is established. The optimization problem of depth interpolation energy model is transformed into a sparse linear equation system. The experimental results show that compared with the popular semi-automatic 2D / 3D method, the proposed method can estimate the depth map with a higher PSNR and enhance the object boundary quality of the depth map.
【作者单位】： 南京理工大学泰州科技学院计算机科学与技术系;宁波工程学院电子与信息工程学院;
【基金】：国家自然科学基金资助项目(61170200) 浙江省自然科学基金资助项目(LY16F010014)
【分类号】：TP391.41
，
本文编号：2247684
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2247684.html
