Separating non-stationary from stationary scene components in a sequence of real world TV-images
IJCAI'77 Proceedings of the 5th international joint conference on Artificial intelligence - Volume 2
| Hi-index | 0.10 |
This paper is concerned with the initial segmentation step of locating the dominant discontinuities in a variety of image ensembles. Using vector representation for the image ensembles, we model the image vectors over a small neighborhood as an outcome of a linear transformation operating on the image vector at the center of the neighborhood. It is shown that such a modelling leads to a two-stage discontinuity detection process. During the first state, a measure of similarity is computed between the neighboring image vectors using the normalized inner product. These similarity values are then combined through two masks to obtain a measure of the discontinuity present at each location. Such a general formulation makes the proposed discontinuity detector capable of handling a variety of images.