A novel parallel approach to character recognition and its VLSI implementation

Character recognition has become more important over the years in areas of document processing, language translation, electronic publication, office automation, etc. Good feature extraction is critical to the success of character classification, especially for very large character sets. In this paper, a parallel VH2D (Vertical-Horizontal-2Diagonal) method and its VLSI implementation are proposed. In the VH2D method, projections on character images are made from four directions: vertical, horizontal and two diagonals (45 and 135 degrees), which produce four subfeature vectors for each character. Four subfeature vectors are transformed according to the central point of the character before they are combined into a complete feature vector for a given character. In this research, for the experiment, the character dictionary consists of 3000 feature vectors of the character set. The experimental results indicate that all the input characters in the dictionary are correctly classified and all the characters outside the dictionary are rejected. The proposed approach contains extensive pipelining and parallelism. The time complexity of the proposed algorithm is O(N) instead of O(N-2) when a uniprocessor is used, where N is the dimension of the digitized image of the input character. A study on a simple VLSI architecture composed of four linear arrays of processing elements (PEs) for the proposed VH2D approach is also presented.