A generalized symmetry transform evaluates evidence of local symmetry, yet it has no selective attention capability for corner angles. Therefore, a symmetry transform is proposed that can recognize symmetric features tuned to specific corner angles. Experimental results with synthetic and real imagery confirm the effectiveness of the proposed transform in discriminating polygonal objects.

In this paper, we develop a block matching method for fast rotation of binary images. We define coarse and fine blocks to extract bit patterns of an original image and calculate their predrawn mapping patterns (PMPs) using the given rotation angle. The traditional calculation for rotation of images can be thus replaced with simple matching of bit patterns of blocks and drawing of their two dimensional (2-D) PMPs at the output plane. A scheme of overlapping blocks is also used to solve problems of hole generation and topology variation usually occurring in rotation. Experimental results demonstrate that our proposed method performs best in terms of rotation speed compared to other algorithms.

Describes the combination of multiple feature sets for multiple databases. We adapt MLPs as efficient feature set generators from the original input feature sets. The generated feature sets by MLPs are then coupled into new feature space that is quite useful for combination of multiple feature sets as well as accommodation of multiple partitioned databases. For our experiments four feature sets and three databases were used. Experimental results showed that the proposed feature set combination provides excellent classification performance and can be applied to general multiple feature spaces and database environments.

A hierarchical neural network was studied for large scale pattern recognition problems such as multilingual character recognition. The hierarchical neural network is made up of preclassification and fine recognition modes. The authors constructed clusters for the preclassification mode using self-organizing feature map (SOFM) learning and performed modifying steps for reducing the number of clusters. The clusters contain patterns that have similar characteristics. They adopted multilayer perceptron (MLP) networks to the corresponding clusters for the fine recognition mode. They also proposed the use of false-alarming nodes in the output layer of the MLP network, which could be constructed on error-prone negative examples quite similar to the patterns of the selected cluster but actually belonging to different nearby clusters through SOFM's topology-preserving mapping. The proposed system could be successfully adopted for recognizing the large number of printed Korean/Chinese character databases as well as the IRIS database.

The authors propose the adaptive window algorithm with four-direction sizing factors (AWA-FSF) for robust correlation-based tracking in complex cluttered environments. When the size and shape of a moving object have been changed, the correlator often accumulates walk-off error. The success of correlation-based tracking largely depends on choosing a suitable window size and position and thus transferring the proper reference image to the next frame. Since the AWA-FSF is capable of adjusting a reference image size more rapidly and properly, one can minimize the influence of complex background and clutter. In addition, one can finely tune the center point of the reference image repeatedly after the main tracking process. Thus one has increased stability and reliability of correlation-based image tracking. They tested the performance of the AWA-FSF using 45 real image sequences made of over 3400 images.