Recovery of motion parameters from distortions in scanned images
Recovery of motion parameters from distortions in scanned images
复制标题
从扫描图像的失真中恢复运动参数
DOI：
--
复制
发表时间：
1997
期刊：
影响因子：
--
通讯作者：
J. Mulligan
中科院分区：
文献类型：
--
作者：
J. Mulligan
当有目标运动时，扫描的图像，如扫描激光检眼镜(SLO)产生的图像，会显示失真。这是因为对应于不同图像区域的像素是顺序获取的，因此本质上是不同快照的切片。虽然这些失真给图像配准算法带来了问题，但它们对于在高于帧速率的时间频率上恢复目标运动参数具有潜在的有用作用。Stetter、Sendtner和Timberlake测量了SLO图像中的大失真，以恢复快速水平扫视眼球运动的时间进程。在这里，这项工作的扩展目标是自动恢复二维的微小眼球运动。使用低维参数描述对帧间隔期间的眼睛位置进行建模，该低维参数描述进而用于生成参考模板的预测失真。然后使用归一化互相关将输入图像配准到失真模板。然后改变运动参数，并重新计算相关性，以找到使相关性的峰值最大化的运动。采用双二次插值法以亚像素精度确定相关极大值的位置和值，得到优于1角分的眼部位置分辨率。使用实际的SLO图像和模拟图像对该运动参数估计方法进行了测试。运动参数估计也可以应用于单独的视频线，以减少接近实时系统的流水线延迟。
Scanned images, such as those produced by the scanning-laser ophthalmoscope (SLO), show distortions when there is target motion. This is because pixels corresponding to different image regions are acquired sequentially, and so, in essence, are slices of different snapshots. While these distortions create problems for image registration algorithms, they are potentially useful for recovering target motion parameters at temporal frequencies above the frame rate. Stetter, Sendtner and Timberlake measured large distortions in SLO images to recover the time course of rapid horizontal saccadic eye movements. Here, this work is extended with the goal of automatically recovering small eye movements in two dimensions. Eye position during the frame interval is modeled using a low dimensional parametric description, which in turn is used to generate predicted distortions of a reference template. The input image is then registered to the distorted template using normalized cross correlation. The motion parameters are then varied, and the correlation recomputed, to find the motion which maximizes the peak value of the correlation. The location and value of the correlation maximum are determined with sub-pixel precision using biquadratic interpolation, yielding eye position resolution better than 1 arc minute. This method of motion parameter estimation is tested using actual SLO images as well as simulated images. Motion parameter estimation might also be applied to individual video lines in order to reduce pipeline delays for a near real-time system.