A Connectionist Approach to the Correspondence Problem in Computer Vision

A problem which often arises in computer vision is that of matching corresponding feature points within images. The correspondence is complicated by the fact that the feature sets are not only randomly ordered but have also been distorted by some transformation whose parameters are usually unknown. If these parameters are completely unknown, then all n! permutations must be compared. A neural computational method is proposed to cope with this combinatorial problem. The method is applied to 2D point correspondence and 3D-to-2D point correspondence, and is also extended to a Boltzmann machine implementation which is less sensitive to local minima, regardless of initial conditions.

[1]  Martin A. Fischler,et al.  Computational Stereo , 1982, CSUR.

[2]  Wei-Chun Lin,et al.  A connectionist approach to multiple-view based 3-D object recognition , 1990, 1990 IJCNN International Joint Conference on Neural Networks.

[3]  Thomas S. Huang,et al.  Uniqueness and Estimation of Three-Dimensional Motion Parameters of Rigid Objects with Curved Surfaces , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[4]  Jake K. Aggarwal,et al.  On the computation of motion from sequences of images-A review , 1988, Proc. IEEE.

[5]  Adam Krzyzak,et al.  Motion estimation based on point correspondence using neural network , 1990, 1990 IJCNN International Joint Conference on Neural Networks.

[6]  Takafumi Miyatake,et al.  A position recognition algorithm for semiconductor alignment based on structural pattern matching , 1989, IEEE Trans. Acoust. Speech Signal Process..

[7]  Hiroshi Sakou,et al.  An Algorithm for Matching of Distorted Waveforms Using a Scale-Based Description , 1988, MVA.

[8]  Rodney A. Brooks,et al.  Model-Based Three-Dimensional Interpretations of Two-Dimensional Images , 1981, IEEE Transactions on Pattern Analysis and Machine Intelligence.