Two-stage box connectivity algorithm for optical character recognition

The box connectivity approach (BCA) uses a regular grid to partition the character bitmap into n × n boxes. The bitmap is then represented by a graph whose vertices and edges correspond to the boxes and their connectivity. The adjacency matrices of the graphs are represented by two binary matrices of lower size: one for the vertical and one for the horizontal connections. A third binary matrix is used to represent pixel densities in each of the boxes. Hamming distances are used for multicriterion classification of the corresponding binary vectors: only noninferior (Pareto optimal) vectors are retained. Size of the character image and first order Markov chain model of the adjacent characters are used to disambiguate noninferior characters in the second stage of the algorithm