Image Processing Reference
In-Depth Information
detail representation. This representation is very useful to shape characterisation. A curvature
scale space has been developed (Mokhtarian, 1986) and (Mokhtarian, 1992) to give a
compact way of representing shapes, and at different scales, from coarse (low-level) to fine
(detail).
Another approach to motion estimation has considered the
frequency domain
(Adelson,
1985) (yes, Fourier transforms get everywhere!). For a further overview of dense optical
flow see Bulthoff (1989) and for implementation see Little (1988). The major survey
(Beauchemin, 1995) of the approaches to optical flow is rather dated now, but the authors
did produce freely available software
(ftp://csd.uwo.ca/pub/vision)
for the
techniques that they also compared in a performance appraisal (Barron, 1994). Such an
(accuracy) appraisal is particularly useful in view of the number of ways there are to
estimate it. The nine techniques studied included the differential approach we have studied
here, a Fourier technique and a correlation-based method. Their conclusion was that a local
differential method (Lucas, 1981) and a phase-based method (Fleet, 1990) offered the most
consistent performance on the datasets studied. However, there are many variables, not
only in the data but also in implementation, that might lead to preference for a particular
technique. Clearly, there are many impediments to the successful calculation of optical
flow such as change in illumination or occlusion (and by other moving objects). In fact,
there have been a number of studies on performance, e.g. of affine flow in Grossmann
(1997). More recently, a thorough analysis of correlation techniques has been developed
(Giachetti, 2000) with new algorithms for sub-pixel estimation. One of the more recent
studies (Liu, 1998) notes how the more recent developments have been for fast or accurate
techniques, without consideration of the trade-off between these two factors. The study
compared the techniques mentioned previously with two newer approaches (one fast and
one accurate), and also surveys real-time implementations that include implementation via
parallel computers and special purpose VLSI chips.
4.9
References
Adelson, E. H. and Bergen, J. R., Spatiotemporal Energy Models for the Perception of
Motion,
Journal of the Optical Society of America
,
A2
(2), pp. 284-299, 1985
Apostol, T. M.,
Calculus
, 2nd Edition,
1
, Xerox College Publishing, Waltham, 1966
Asada, H. and Brady, M., The Curvature Primal Sketch,
IEEE Trans. on PAMI
,
8
(1), pp. 2-
14, 1986
Barnard, S. T. and Fichler, M. A., Stereo vision, in
Encyclopedia of Artificial Intelligence
,
New York: John Wiley, pp. 1083-2090, 1987
Barron, J. L., Fleet, D. J. and Beauchemin, S. S., Performance of Optical Flow Techniques,
International Journal of Computer Vision
,
12
(1), pp. 43-77, 1994
Beauchemin, S. S. and Barron, J. L., The Computation of Optical Flow,
Communications
of the ACM
, pp. 433-467, 1995
Bennet, J. R. and MacDonald, J. S., On the Measurement of Curvature in a Quantised
Environment,
IEEE Trans. on Computers
,
C-24
(8), pp. 803-820, 1975
Bergholm, F., Edge Focussing,
IEEE Trans. on PAMI
,
9
(6), pp. 726-741, 1987
Bovik, A. C., Huang, T. S. and Munson, D. C., The Effect of Median Filtering on Edge
Estimation and Detection,
IEEE Trans. on PAMI
,
9
(2), pp. 181-194, 1987
Bulthoff, H., Little, J. and Poggio, T., A Parallel Algorithm for Real-Time Computation of
Optical Flow,
Nature
,
337
(9), pp. 549-553, 1989
