Visible models for interactive pattern recognition
Pattern Recognition Letters
Tools for monitoring, visualizing, and refining collections of noisy documents
Proceedings of The Third Workshop on Analytics for Noisy Unstructured Text Data
Notes on contemporary table recognition
DAS'06 Proceedings of the 7th international conference on Document Analysis Systems
| Hi-index | 0.00 |
Almost all operational visual pattern recognition systems require some human assistance either at the beginning or at the end. The motivation for interactive pattern recognition is simply that it may be more effective to make parsimonious use of human visual talent throughout the process. We introduce the concept of Computer Assisted Visual InterActive Recognition (CAVIAR). In object classification with CAVIAR, a domain-specific geometrical model, e.g., a set of contours or critical feature points, plays the central role in facilitating the communication (interaction) between human and computer. The key to effective interaction is the display of the automatically-fitted adjustable model that lets the human retain the initiative throughout the classification process. The alternating human and computer steps in the CAVIAR process are modelled as a finite state machine. The computer tries its best to estimate an initial model for the unknown sample and calculate its similarity to the training samples that belong to each class. Representative training pictures are displayed in the order of computer-calculated similarities. The model is also displayed to the user, who can correct it if necessary. Any correction leads to an update of the CAVIAR state, re-estimation of the remaining unadjusted model parameters, and re-ordering of the candidates. A CAVIAR classification is concluded with a final confirmation by the user. We demonstrate the effectiveness and wide applicability of the proposed methodology by implementing two systems: CAVIAR-flower and CAVIAR-face. Evaluation of these two systems on 51 subjects reveals that: (1) CAVIAR can significantly reduce the recognition time compared to the unaided human, and significantly increase the accuracy compared to the unaided machine; (2) human-computer communication through a geometrical model is effective; (3) the CAVIAR system can be initialized with a single training sample per class, but still achieve high accuracy; (4) the CAVIAR system shows self-learning ability and improves with use. CAVIAR is being ported to a mobile hand-held computer as a client connecting to an Internet server. Possible applications to other domains include face, sign, and skin disease recognition.