基于SIFT 匹配的多视点立体图像零视差调整

针对目标跟踪过程中目标可能出现的快速变化和严重遮挡等问题,提出了一种基于新的子空间表示的目标跟踪算法。采用距离不变量对尺度不变特征变换(SIFT)特征点匹配对进行提纯。用提纯后的特征点匹配对,通过线性拟合得到仿射变化参数。在粒子滤波的理论框架下,采用快速的迭代算法,建立目标的主分量(PCA)子空间表示,结合计算得到的仿射变化参数,构造有效的目标观测模型完成跟踪。同时,采用在线学习的方法对SIFT特征点和PCA子空间进行定时更新。大量实验表明,提出的算法能快速有效地完成对姿态和形状剧烈变化的目标的精确跟踪。

Zero-disparity adjustment of multiview stereoscopic images based on SIFT matching

A zero-disparity adjustment method based on SIFT matching was proposed for multiview stereoscopic images using in autostereoscopic display system. First, SIFT was introduced for pixel matching between adjacent views. Then, the result of SIFT matching was filtered by saliency mask which was extracted using frequency-tuned saliency model, and the key-point of disparity control was selected. Finally, the disparity between the neighboring views was computed based on SIFT matching points, and zero-disparity adjustment was conducted based on the principle of disparity control. The disparity of selected key-point was adjusted to zero. Experimental results demonstrate that the proposed method can effectively adjust the disparity of multiview stereoscopic images and generate vivid and comfortable 3D scenes for autostereoscopic display.