Search

A technique for detecting edges directly from a vector-quantized image representation, called codebook edge detection was proposed by Mclean (CVGIP: Graphical Models Image Process. 55, 1993, 48-57). With this method, the edges can be detected by a simple table lookup, a so-called "edge codebook." In this paper, we propose an improved method for codebook edge detection, which is based on a new type of edge codebook. Experimental results provide that much better performance can be gained using this new kind of edge codebook, whose size is even smaller than that of Mclean's.

This paper presents a new method for detection of the moving trajectory of a pixel-sized target from a time sequence. First, a noise reducing algorithm based on a quasi-continuity-filter is used to filter most of the noise pixels out effectively. Then, the properties of a trajectory are embedded in a cost function and the trajectory detection problem becomes the problem of minimizing the given cost function. Finally, experiments show the advantage of this method for detection of arbitrary target trajectories in high noise environments without the major drawback of limited detectability within the predefined trajectory set or massive computation that most other algorithms suffer from.

In the case of complex background, the target cannot be extracted perfectly. Existing methods for object recognition are not effective any more. A dynamic algorithm for tank recognition in a complex background by moment invariant is proposed in this paper. It extracts and identifies the object simultaneously. Experiment results have shown that this algorithm can identify tank objects with low contrast in a complicated scene successfully.

A new continuity filter method for detecting small moving targets with unknown trajectories in image sequences of a high noise environment is presented.