| 摘要 |
PURPOSE:To perform recognition in consideration of the information of a context by outputting the candidate of a symbol to show the information to be recognized together with confidence and selecting base on the output value of the confidence and node in the case of many candidates. CONSTITUTION:At a recognizing part 2, the pattern of the character image, voice, etc., inputted from an input part l is analyzed and the combination of the candidate symbol and the confidence is outputted by a candidate output part 7. A selecting part 8 judges the candidate symbol from the output value and the confidence in a network part 9 and selects and outputs an optimum symbol. In this case, the network part 9 is constituted of a neuron circuit network to perform association. The output value goes up and down by the teacher signal that a user for the symbol outputted by the device receives or rejects. Then, the output value of the node to show the symbol deeply related to the topic inputted up to then becomes higher. Thus, in consideration of not only the information of the pattern but also the information of the context in which the recognition object appears, the suitable candidate can be drawn. |
