Keyword spotting in unconstrained handwritten Chinese documents using contextual word model
Image and Vision Computing
| Hi-index | 0.00 |
Abstract: Making a large database of electronic documents from paper documents has left a tremendous problem. In order to search the database for an image document, it is necessary for general electronic filing systems to convert the document into texts using OCR. However, the system cannot retrieve documents that do not contain correct character codes. We had before proposed a document retrieval method that reduces false drops and false alarms by using the "shape-feature" technique that describes the outline of the character's shape. We have now applied this method to handwritten Japanese documents. Experimental results reveal that our method has a high recall rate of 88.8% compared to the conventional methods (69.2%: text matching, 78.3%: candidate matching).