A connectionist model for graytone thinning
J. Basak, Nirmalya Pal, Pritesh Patel · 2002
Multilayered perceptron (MLP), capable of generating nonlinear decision boundaries, can be used for designing templates or convolution operators for image thinning. Given a parallel thinning algorithm, the set of rules specifying the deletion conditions of a pixel can be learnt using the MLP. The weights of the links in that case represent the corresponding template weights of the convolution operator. The objective of using MLP is to develop a general computational framework where given any parallel thinning algorithm for two-tone images, we can have a connectionist model for both two-tone and gray-tone image thinning. Our strategy is as follows: train an MLP with two-tone images and then use it for graytone images with some additional normalization operation on the input images. Due to the generalization ability of MLP, we expect to get some reasonable output for graytone images also.