| 摘要 |
PURPOSE:To automatically prepare a dictionary for pattern recognition by separating a feature space into an area including only positive samples included in a learning category and an area including only negative samples which are not included in the learning category. CONSTITUTION:In a learning phase, a feature extraction part 1 extracts plural features from an input sample pattern and a threshold value found by a threshold value learning part 2 is stored in the learning dictionary 3. In a recognition phase, the feature extraction part 1 extracts feature quantities from an unknown input pattern and a threshold value comparison part 4 compares their values with the threshold value in the learning dictionary 3 to recognizes a pattern which satisfies all threshold conditions as a corresponding category. When a positive sample included in the category and a negative sample as a pattern which is not included in the category are inputted at the time of learning, the feature space is automatically separated into the area including only the positive sample and the area including only the negative sample and they are stored as the learning dictionary. |
