機器視覺問卷輸入影像辨識研究

Study of the Automatic Recognition for Questionnaire Input by Machine Vision

張玉其(Yu-Chi Chang)

指導教授 ： 蒲永仁

長榮大學/健康科學學院/職業安全與衛生研究所/碩士(2010年)

https://doi.org/10.6833/CJCU.2010.00153

摘要

問卷調查係專家學者採用的重要研究方法之一，在公共安全衛生的各個領域經常利用問卷來進行計量性研究，其歷史悠久且應用廣泛。問卷調查的實施有其嚴謹的步驟，為求分析資料的準確性或代表性，通常希望回收的問卷數量在一定的數目以上，多則數千或數萬份，使得問卷輸入成為整個研究流程的重要關卡。問卷填答大都採用塗滿或勾選空格的方式進行，問卷輸入即可利用人工鍵入或讀卡機讀取後傳輸至電腦檔案完成。人工作業工作量非常大，而讀卡機輸入則需使用專用的卡紙，不但需要委外印刷，而且必須租用讀卡機進行讀卡，成本自然增加。因此問卷的版面設計及紙張材料若無任何限制，研究者亦能在自身研究室完成大量地自動化輸入，即成為本文研究之方向。
本論文藉由圖控式程式語言的撰寫，輸入回收問卷的影像，手動定位空格，即時進行影像分析處理。程式依空格內黑點像素量判斷空格有無填答，首先由實驗分析訂定勾選門檻值為黑色像素佔框內區域面積百分之十。接著對多張打勾和塗抹的圖形進行像素水平投影分析，利用已知的填答結果，訂出像素面積加上投影特徵作為塗抹門檻值。接下來利用上述這些門檻值，令程式實際辨識問卷影像，測試其辨識成功率，實驗結果得到程式辨識率超過99%，且針對單一空格平均辨識時間只需10 ms，單張問卷辨識時間平均為785.5 ms，程式不但具備快速影像資料辨識功能，也為自動化機器視覺問卷輸入系統奠定了良好的基礎。

關鍵字

公共安全衛生 ； 問卷 ； 機器視覺

並列摘要

Questionnaire survey is one of the important research tools for scholars, especially in the areas of public health and social science, to conduct quantitative information collections. The process of a questionnaire surrey usually includes several phases, such as question design, format arrangement, pre-test, administration and feedback. For the representative and precision, the quantity of questionnaires is usually quite large that sometimes can reach thousands; therefore the data collections become one of the key steps during research. One of the traditional data inputs uses manpower by key punching each answer manually into a spreadsheet cell; other more efficient way is to use the card scanners to read those Scantron cards specifically designed for each machine. Manual key-in method takes a lot of manpower and can be time consuming. Errors occur from time to time and sometimes need equivalent manpower to locate them by comparing each other’s result. While the scanner method not only limits people to setup and choose papers, but also costs much higher since consulting a commercial company is inevitable. Due to the inconveniences mentioned above, this study sets out to find an automatic and vast input method so that the future designs for questionnaires are free to paper materials as well as the question formats.
Coded with a visual programming language in this thesis, the developed software is integrated with a CCD device, wherein the questionnaire images are captured, digitized and processed in real time. In other to distinguish the presence of marks in each checkbox, the foreground area percentage was calculated in the first experiment, which determined the corresponding threshold of 10% for a marked checkbox. Next, we analyzed the horizontal projections of the foreground pixels for the known answered checkboxes. The thresholds of the erasing marks were determined by some features therein. Provided with the thresholds mentioned above, the developed software processed the acquired images of the collected questionnaires and found the rate of accuracy over 99%. Moreover, the average processing time took only 10 ms to recognize one checkbox, while it took 785.5 ms to process a single page. The results demonstrate that our software not only possesses the ability of fast image recognition, but also lays a foundation for the automatic questionnaire input system.

並列關鍵字

Public Safety and Health ； Questionnaire ； Machine Vision

參考文獻

[1] H. Al-Yousefi and S. S. Udpa, “Recognition of Arabic Characters,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 14, pp. 853-857, Aug. 1992.

[2] Weihua Huang, Chew Lim Tan, Sam Yuan Sung and Yi Xu, “Vertical Bar Detection for Gauging Text Similarity of Document Images,” Proceedings of Sixth International Conference on Document Analysis and Recognition, pp. 640-644, Sep. 10-13, 2001.

[3] Weihua Huang, Chew Lim Tan, Sam Yuan Sung and Yi Xu, “Word Shape Recognition for Image-Based Document Retrieval,” Image Processings of International Conference, vol. 1, pp. 1114-1117 , Oct. 7-10, 2001.

[4] In-Jung Kim, “Multi-Window Binarization of Camera Image for Document Recognition,” Ninth International Workshop on Frontier in Handwriting Recognition, pp. 323-327, Oct. 26-29, 2004.

[5] National Instruments, IMAQ Vision Concepts Manual, 2003.