Motion Estimation in Digital Angiographic Images Using Skeletons J.Y. Kwak, S.N. Efstratiadis, A.K. Katsaggelos, A.V. Sahakian and B.J. Sullivan Northwestern University, Dept. of Electrical Engineering and Computer Science McCormick School of Engineering and Applied Sciences, Evanston, illinois 60208-3118 S. Swiryn, D.C. Hueter, and T. Frohlich Northwestern University Medical School, Dept. of Medicine, Division of Cardiology The Evanston Hospital, Evanston, Iffinois 60201 Abstract This paper deals with the estimation of the motion field in digital angiographic sequences. An approach is developed according to which each frame is first segmented into a moving object of interest and the background. The original images are converted into binary images by using a Gaussian smoothing filter and thresholding. In the binary images, pixels in moving objects have the value of "1" and pixels in the background have the value of "0" . The moving objects in the binary images are thinned to skeletons with unit width by using the Safe-Point Thinning Algorithm (SPTA) with restriction windows we introduce. Then, a block-matching algorithm is used in estimating the motion for the pixels which belong to the skeleton. This approach to motion estimation results in reduced computations since only binary multiplications need to be performed for determining the match between two blocks. Therefore, an effective searching method is proposed for finding the direction of displacement in successive skeleton frames. Very satisfactory results are obtained by applying the algorithm to 64 x 64.pixel digital angiographic image sequences. 1 Introduction Motion estimation plays an important role in the processing and analysis of sequences of images. Knowledge of the displacement vectors can be used for motion-compensated prediction coding, motion- compensated transform coding and motion adaptive frame interpolation schemes. The motion esti- mation algorithms that have appeared in the literature can be grouped into pel-recursive and block- matching algorithms1'46'8. Pel-recursive algorithms make use of the information from the spatial and temporal difference signals between successive frames and produce an estimate of the dispacement vector in a recursive manner. Block-matching algorithms are basically searching techniques of the displacement vector which satisfies a matching criterion between blocks of the same size in two successive frames. The important underlying assumption of the latter is that, all pixels in a block have the same displacement. In the block-matching algorithms computational reduction can be achieved by simplifying the matching criterion; that is, instead of the normalized cross-correlation function, the mean-squared error1 and the mean absolute difference4 can be used. Furthermore, various methods, e.g., the 2-D logarithmic search1, the three-step search4, and the modified conjugate direction search8 have been proposed to accelerate the search procedure. The objective of this paper is the estimation of the motion vectors in digital angiographic image sequences. Knowledge of the motion field can be incorporated into image enhancement strategies. One such enhancement would be the improved depth-perception, where motion information would be used to identify different planes in a network of arteries. A second enhancement technique would use a model of the motion to predict the location of structures that become temporarily obscured due to merging with other structures along the same imaging path or to poor signal-to-noise ratio. The latter case may 32 / SPIE Vol. 1396 Applications of Optical Engineering: Proceedings of OE/Midwest '90