Hough transform network: a class of networks for identifying parametric structures

摘要

A class of structure seeking neural networks is presented which are capable of learning parametric structures under unsupervised mode. The functionality of the class of networks is analogous to that of the classical Hough transform, one of the most widely used algorithms in visual pattern recognition. However, the present class of networks provide a much more efficient representation with a highly reduced storage space, capability of quantifying the impreciseness in the input, and ability to handle sparse data sets. The effectiveness of the network and its newly defined learning rules is demonstrated on different data sets under noisy conditions.