查詢結果分析
來源資料
頁籤選單縮合
題名 | On Speeding Radical-Based On-Line Chinese Character Recognition Through the Directional Matching of the First and Last Segments=使用前後筆段的方向比對用於加速以字根為基礎的線上手寫中文文字辨認系統 |
---|---|
作者 | 周國森; 張保忠; 徐克華; 陳玲芬; 許超智; 王亮盛; | 書刊名 | 電信研究 |
卷期 | 28:5 1998.10[民87.10] |
頁次 | 頁731-747 |
分類號 | 312.74 |
關鍵詞 | 線上手寫中文文字辨認系統; OLCCR; On-line Chinese character recognition; Radical; |
語文 | 英文(English) |
中文摘要 | 字根是中文文字組成的基本要件之一,字根的使用可以大幅降低在細分類中所要 比對的字數,從而加速整個辨認系統,在線上手寫中文文字的識別上,字根除了以上的優點 之外,更可以吸收筆順與連筆上的變動,更可以使識別系統得以處理因同義字根介入整體字 形中所造成的字形變動,在字根的辨認上,首先可以使用輸入的筆劃數來界定比對的範圍, 在完成字根辨認後再施行字形的比對,整個辨認系統的平均辨認時間為 0.756 秒╱字( PC 386-33 ),而其平均辨認率為 96.63%, 若考慮前十位候選字則其累積辨認率則可達 99.02%, 為了更形提升系統的辨認速度本文中題出一種字根擷取的方法,使整個辨認系統 的平均辨認時間為 0.454 秒╱字( PC 386-33 ),而其平均辨認率為 96.76%,若考慮前 十位候選字則其累積辨認率則可達 99.07%,若在字根辨認之後,將距離值太大的字根予以 去除,則整個辨認系統的平均辨認時間為 0.427 秒╱字。 |
英文摘要 | The beauty of Chinese characters is that most of them possess a radical with a meaning. Our system is motivated by this characteristic owned in Chinese characters which leads us to a radical-based candidate selection approach performed before similarity measurement. The searching ranges of the radicals can be basically decided by the number of stokes in the input script. By doing so, the speed of our system achieves 0.756 seconds in average per character on a PC using Intel 38633 CPU and the system performance for the input script to be recognized in the first rank is 96.63% and in the first 10 candidates is 99.02%. Furthermore, to speed up the inspection of radical candidates, we developed a radical extraction algorithm to reduce the searching scope of radical templates. By applying this algorithm, the recognition rate improves to be 96.76% for the first rank and 99.07% for the first 10 candidates with the speed of 0.454 seconds in average per character. Since some of the candidate radicals have very high matching cost, we can even further ameliorate the speed by discarding those unlikely candidate radicals. By this way, the performance can be maintained, yet the time spent for a character to be processed is 0.427 seconds in average. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。