Font Size: a A A

Chinese Postal Address Recognition

Posted on:2005-06-18Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z L LouFull Text:PDF
GTID:1118360185995674Subject:Artificial intelligence and pattern recognition
Abstract/Summary:PDF Full Text Request
To improve the performance of Chinese post adress sysytem, the weak parts of handwritten character manuscript recognition are focused on in my research. Several key components of manusript disposal have been studied in this paper, such as binarization, segmentation, recognition, post-processing, and so on. The research results have been integrated into a practical application system as well.Following are main contributions of this paper,1. A kind of region binarization is proposed. The proposed methods can obtain a balance on whole binarization and local binarization. A good adaption is held on complicated documents. At the same time, a less time is consumed.2. According to the shape features of handwritten characters, competitive dynamic program is adopted to realize an algrithm for segmentation. The accuracy of the algrithm can reach 80 percent. In this algrithm, Viterbi algrithm is adopted to segment characters, and dynamic program is adopted to seek optimal segmentation path.3. A new fine classification algrithm is presented, which was applied into unconstrained offline handwritten numerals in my former work and based on wavelet transformation and local Fourier transformation. The presented algrithm is integrated into handwritten character recognition engine of Hanwang Company. The accuracy of the first candidate is improved by 3.8%.4. A new lexicon-driven segmentation and recognition of handwritten character strings for Chinese address reading is proposed. By multi-segmentation path, multi-recognition results and address lexicon, the highest weight path is choosed for the last recognition results. Excellent results are demonstrated by experiments.5. A fuzzy match algorithm of Chinese character strings is proposed for post-processing. In the case of some errors occurred in segmentation and recognition, sensitive address string, which includes at least five characters, was extracted from envelopment image. In experiments, the detection rate is 95%; false reject rate only is 15%.6. Based on the techniques mentioned above, a mail search-system is designed. In present, the aim of first stage of this project is reached. The second stage of this project, which applied into practice, will also be finished soon.
Keywords/Search Tags:binarization, post-processing, handwritten character recognition, mail search
PDF Full Text Request
Related items