A neural network shape recognition system with Hough transform input feature space

The Hough transform (HT) has been used as an efficient method for straight line and parametric shape detection because of its robustness against noise and occlusion. The authors have developed a system which uses a neural network (NN) to recognize data extracted from the Hough space (HS). The system comprises three stages: the first stage carries out median filtering, histogram analysis, binarizing, Sobel edge detection and thinning to get the thinned outline of the shape The second stage HTs the edge image, eliminates the translational, scaling and rotational parameter dependency of the HS and extracts the feature vectors from the HS. The final stage is a NN which takes in the feature vectors as input and performs learning and recognition tasks