A connectionist system for learning and recognition of structures: Application to handwritten characters

Hough transform network: learning conoidal structures in a connectionist framework

A two-layer neural-network model is designed which accepts image coordinates as the input and learns the parametric form of conoidal shapes (lines/circles/ellipses) adaptively. It provides an efficient representation of visual information embedded in the connection weights and the parameters of the processing elements. It not only reduces the large space requirements of the classical Hough transform (HT), but also represents parameters with a higher precision. The performance of the methodology is compared with other existing algorithms and has been found to excel over those algorithms in many cases.

Integrated feature analysis and fuzzy rule-based system identification in a neuro-fuzzy paradigm

Most methods of fuzzy rule-based system identification (SI) either ignore feature analysis or do it in a separate phase. This paper proposes a novel neuro-fuzzy system that can simultaneously do feature analysis and SI in an integrated manner. It is a five-layered feed-forward network for realizing a fuzzy rule-based system. The second layer of the net is the most important one, which along with fuzzification of the input also learns a modulator function for each input feature. This enables online selection of important features by the network. The system is so designed that learning maintains the nonnegative characteristic of certainty factors of rules. The proposed network is tested on both synthetic and real data sets and the performance is found to be quite satisfactory. To get an "optimal" network architecture and to eliminate conflicting rules, nodes and links are pruned and then the structure is retrained. The pruned network retains almost the same level of performance as that of the original one.

Learning Hough transform: a neural network model

A single-layered Hough transform network is proposed that accepts image coordinates of each object pixel as input and produces a set of outputs that indicate the belongingness of the pixel to a particular structure (e.g., a straight line). The network is able to learn adaptively the parametric forms of the linear segments present in the image. It is designed for learning and identification not only of linear segments in two-dimensional images but also the planes and hyperplanes in the higher-dimensional spaces. It provides an efficient representation of visual information embedded in the connection weights. The network not only reduces the large space requirement, as in the case of classical Hough transform, but also represents the parameters with high precision.