Research Of Handwritten Chinese Characters Recognition Based On Generalized Characteristic Feedback Method

Handwritten Chinese character recognition has become a research hotspot in the field of artificial intelligence and pattern recognition because of its wide range of application prospects. But it has been considered the most difficult to conquer in the field of Chinese character recognition for its complexity. The human psychology cognition has been considered a process which contains feedback in the psychology and physiology. The traditional Chinese character recognition system is basically a open-loop system, the one-way recognition process does not fully meet the human recognition process. A recognition of organic whole with feedback and close contact between the various modules is established in the paper. Try to further improve the system performance and improve accuracy and false reject rate in Handwritten Chinese character recognition process.The following are the major contents in the thesis:1. Handwritten Chinese characters recognition system with feedback. A handwritten Chinese characters recognition system with three parallel-coupled structure is established by imitating the human brain recognition contains a feedback process. The system includes decision-making layer, recognition layer and feedback layer. And its operating mechanism has been given. A handwritten Chinese characters recognition system with the feedback correction mechanism for different recognition error results is established.2. Multi-modal qualitative and quantitative recognition. The system extracts kinds of global features to constitute the object's multi-modal feature model which integrates handwritten Chinese characters'quantitative and qualitative features.On this basis, the system realizes the multi-modal qualitative recognition. A decision-making mechanism is established for choosing the best first round recognition method. And the handwritten Chinese characters are recognized by the recognition method.3. General characters recognition error's characterization and acquirement. Based on error analysis to recognition results, three kinds of general characters recognition errors are designed in the paper and the error analysis method is also been given which integrates the three errors'quantitative and qualitative features at the same time. A credibility evaluation index system and a feedback correction decision-making mechanism of the recognition results are established by analyzing the three errors'type and value.4. Choose three types and total of 400 handwritten Chinese sample characters in SCUT-IRAC to validate the feedback recognition system. The experiment results have shown that that method is effective. It increases accuracy and reduces false reject rate of handwritten Chinese characters recognition at the same time.