Shields M E, Farison J B, Miller J W
Department of Elecrical Engineering and Computer Science, University of Toledo, Ohio 43606, USA.
Biomed Sci Instrum. 1995;31:195-200.
This paper explores the use of an image sequence processing algorithm, called the simultaneous diagonalization (SD) filter, which can be effectively applied to long noisy image sequences. This filter was developed to filter a spatially invariant image sequence to form one new image in which a desired feature is enhanced and one or more undesired features (and noise) are suppressed in the filtered image. This filtering technique, applied to a long noisy image sequence, can be used to achieve significant data compression for image storage and provide surprisingly good enhanced image reconstructions. For this investigation, SD filtering is applied to a temporal image sequence, a renogram, with compression of a very noisy 180-image sequence to a 4-image set. The renogram, a nuclear medicine technique, was chosen due to its low signal-to-noise ratio over a long image sequence. Before the application of the SD filter, classical image processing techniques, median and averaging filtering, are used as a preliminary method to reduce the image sequence noise content. Compared to any of the images in the original image sequence, the reconstructed images are remarkably good. The SD filter with prefiltering, thus, can collect information distributed over a 180-image temporal sequence with low signal-to-noise ratio.
本文探讨了一种名为同步对角化（SD）滤波器的图像序列处理算法的应用，该算法可有效应用于长的噪声图像序列。开发此滤波器是为了对空间不变的图像序列进行滤波，以形成一幅新图像，其中所需特征得到增强，而一个或多个不需要的特征（以及噪声）在滤波后的图像中被抑制。这种滤波技术应用于长的噪声图像序列时，可用于实现图像存储的显著数据压缩，并提供出奇好的增强图像重建效果。在本次研究中，SD滤波应用于一个时间图像序列——肾图，将一个噪声很大的180幅图像序列压缩为一个4幅图像的集合。选择肾图这种核医学技术是因为其在长图像序列上的信噪比很低。在应用SD滤波器之前，使用经典图像处理技术——中值滤波和均值滤波作为减少图像序列噪声含量的初步方法。与原始图像序列中的任何一幅图像相比，重建后的图像都非常好。因此，带有预滤波的SD滤波器能够收集分布在一个信噪比很低的180幅图像时间序列中的信息。
