K. Takahashi, H. Amanuma, K. Tanaka, and O. Hanai (Japan)
Pattern recognition, classification, clustering, sampling, prediction, candidate table.
This paper presents a new precise clustering prediction classification method for recognition. A prediction classification based on clustering classification followed by a prediction classification is proposed to get fast and precise processing. Character groups are decided by the maximum distance clustering. Only one dictionary for each group is used for clustering classification to get fast processing. Instead of sample input, we used the higher rank category dictionary in prediction classification. The higher rank category is decided by the former clustering classification. After clustering classification, prediction classification take place by the higher rank dictionary input image based on clustering classification. We obtained twice speed compared to the classification without clustering. And, the decreased number of candidates is a third of the hierarchical clustering classification result.
Important Links:
Go Back
