一种二维图像序列的深度图像生成方法

摘要

This paper proposes a depth map generation method based on motion object. First, the paper uses improved cumulative mean square deviation algorithm to extract background model, then background subtraction method is used to obtain motion object figure. After artificially drawing the depth map of the background model, with motion object figure of each frame, the depth maps of the image sequence are automatically synthesized, which can be directly used in two-dimensional (2D) to three-dimensional (3D) conversion. Experimental results show that compared with conventional depth map generation methods, depth map obtained by this method is true and reliable, thus more comfortable for 3D visualization.%针对二维图像序列提出一种基于运动对象的深度图像生成方法.采用改进的均方差累加算法提取背景模型,并利用背景差分法提取运动对象图形,将人工绘制的背景模型的深度图像,结合每帧运动对象图形深度赋值,自动合成用于二维视频到三维视频转换的图像序列的深度图像.实验结果证明,相对于传统的仅仅依靠计算机视觉获取深度图像的方法,它获得的深度图像,不仅画面的深度信息真实、可靠,而且转换后的三维场景更立体化.