The problem of classifying planar shapes which are only partially specified due to obscuration or segementation error is considered. The boundaries are modeled as circular autoregressive (CAR) processes. The CAR descr...
详细信息
ISBN:
(纸本)0818606339
The problem of classifying planar shapes which are only partially specified due to obscuration or segementation error is considered. The boundaries are modeled as circular autoregressive (CAR) processes. The CAR descriptors are defined from the boundary model and are used to classify the shapes. The maximum-likelihood (ML) estimator of the CAR des criptors is derived first for the complete boundary case and is then extended to the case of incomplete boundaries. The obscured boundary process is modeled as the product of the complete boundary process and a binary-valued (0,1) obscuring process. This model of the obscured boundary process is asymptotically stationary. It is used in extending the ML estimator to include the case of partially obscured boundaries. A classification experiment is performed using four classes of synthetically generated boundary shapes. The classification results obtained using the extended ML estimates are shown to be better than those obtained using the least-squares estimates computed from linearly interpolated boundaries. The extended ML estimator is then demonstrated on real aircraft boundaries.
The cubic polynomial is analyzed and its translation invariant parameters are derived. These translation invariant parameters are scale and contrast and are related to the horizontal and vertical distance between rela...
详细信息
ISBN:
(纸本)0818606339
The cubic polynomial is analyzed and its translation invariant parameters are derived. These translation invariant parameters are scale and contrast and are related to the horizontal and vertical distance between relative extrema of the cubic. The implementation details of the facet model second directional derivative zero-crossing edge detector described previously are then given in terms of the translation invariant parameters. A variety of results are shown for a noiseless sample function having different kinds of discontinuities. Finally, a comparison of the zero-crossing of Laplacian, a popular Mexican hat edge detector, shows that regardless of the standard deviation of the Gaussian, zero crossing slope thresholds which are too small yield some falsely detected edges, and zero crossing slope thresholds which are too large yield some misdetected edges and some incorrectly placed ones. Furthermore, in contrast to the facet edge detector, there is no zero-crossing slope threshold for the Mexican hat edge detector which can provide perfect edge detection on the given noiseless sample function.
暂无评论