| 摘要 |
PURPOSE:To allow reading of thousands of hand-written KANJI (Chinese character) groups simply and easily by constituting a feature extracting part in a character reading device with an edge feature extracting part, a cell feature extraction cell space circuit, edge feature memory and control parameter memory. CONSTITUTION:Character patterns to be observed are entered an pretreatment part B which converts them into an easy-handling pattern by simple normalization, etc., and outputted as a pattern on a lattice surface in which respective lattice points take binary and multiplex values. The normalized pattern enters a feature extracting part C, and features of a stroke, an intersection, an endpoint, etc., of a character are extracted. The feature extracting part C consists of an edge feature extracting part, a cell feature extraction cell space circuit, edge feature memory and control parameter memory. The edge feature extracting part space-differentiates at each lattice point of a character pattern, and extracts a directional vector having the direction and magnitude (edge features). A parameter in the direction of an edge is normally quantized in the eight direction. Edge features determined by an input pattern are decided and its data are stored. Cell features show properly the general features of a character figure; therefore they demonstrate effectiveness at the time of reading of hand-written KANJIs, etc.
|
