Recognition of image patterns generated by a stochastic object-replacement model

We discuss targets in images formed from a pattern of partial occlusions. The targets vary between examples. It is shown that a new scale-space method is substantially more effective than, for example, normalised cross-correlation for distinguishing such a target from background and estimating its position. We believe such targets represent a good model of many objects.

[1]  J. Andrew Bangham,et al.  Multiscale recursive medians, scale-space, and transforms with applications to image processing , 1996, IEEE Trans. Image Process..

[2]  J. Andrew Bangham,et al.  A comparison of linear and nonlinear scale-space filters in noise , 1996, 1996 8th European Signal Processing Conference (EUSIPCO 1996).

[3]  Pierre Chardaire,et al.  Multiscale Nonlinear Decomposition: The Sieve Decomposition Theorem , 1996, IEEE Trans. Pattern Anal. Mach. Intell..

[4]  J. Andrew Bangham,et al.  Multiscale median and morphological filters for 2D pattern recognition , 1994, Signal Process..

[5]  J. Andrew Bangham,et al.  Morphological scale-space preserving transforms in many dimensions , 1996, J. Electronic Imaging.

[6]  J. Andrew Bangham,et al.  Scale-space from nonlinear filters , 1995, Proceedings of IEEE International Conference on Computer Vision.

[7]  J. A. Bangham,et al.  Multiscale median and morphological filters used for 2D pattern recognition , 1993 .