以影像處理方法驗證國家考試答案卡檢測準確性

Abstract

國內的各級考試除了作文外，各學科計分大部份皆採用劃記答案卡的方式：應試者依據題目在答案卡適當的題號答案區中劃記，應試中心則利用光學讀取設備(Optical Mark Reader)，讀取應試者答案卡上劃記的資訊，以作為計分方式。比起過去用人工判讀，以OMR讀取答案卡計分，的確具有快速、凖確等優勢。 然而，要進行OMR的判讀，需有多方條件的配合，例如：答案卡的印刷、裁切、紙張的品質都必須符合規範，否則在在都會影嚮答案卡判讀的準確度，且國內甚少有專文針對國家考試的答案卡計分方式的準確性進行探討。 本研究提出以影像處理的方式來驗證OMR判讀的正確性，選擇以考選部過去360,702萬份考生答案卡為研究樣本，將答案卡以掃描器把答案卡數位化後，進行答案卡辨識，並將辨識結果與考選部原始數據進行比對。 本論文的主要貢獻如下：第一，利用影像辨識方法找出目前國家考試所使用的OMR掃描的準確率上限值。第二，提出一個可行的影像辨識方法作為輔助OMR的工具。第三，對於用影像辨識技術來取代OMR提出了一個初步可行的方法。 實驗的結果顯示OMR的掃描正確率的上限值為百萬分之8.31，以OMR做為答案卡的檢測工具是值得信賴的。

Currently all levels examination other than composition, mostly utilize marker on answer card to take the exam: Based on the subject, candidate mark appropriate answer on answer card's designated area. State Examination Center then use Optical Mark Reader to read candidates answer as ways of counting scores. Compared to the past manual interpretation of reading the answer card, OMR scoring present the advantages of both speed and accuracy.

In order to utilize OMR interpretation of result properly, several criteria must be meet: a printed answer card that meet cutting, paper quality and ink requirements. Currently there are few published articles that discuss the method which State Examination Center is using and the accuracy of score card counting. On our study, we propose image processing of way to verify the result of accuracy on OMR interpretation. We chose past exam result of 3.60702 billion examinee's answer card as our research sample. Digitalizing these score cards by using scanner then compare our result with the original examination department data. The main attributes of this paper are as follows: First, the usage of image recognition method to identify the limitation on the accuracy of current state exams OMR scanner. Second, propose a feasible method of image recognition to assist OMR. Third, present way to utilize image recognition technology to replace OMR.

Our test result has shown that, the accuracy of OMR scanning upper limit is 8.31 parts per million. Which indicates that using OMR answer card as detection tool is trustworthy.