Automatic assessment mark entry system using local binary pattern (LBP) and salient structural features

Offline handwritten digit recognition continues to be a fundamental research problem in document analysis and retrieval. The common method used in extracting handwritten mark from assessment forms is to assign a person to manually type in the marks into a spreadsheet. This method is found to be time consuming, not cost effective and prone to human mistakes. Thus, a number recognition system is developed using local binary pattern (LBP) technique to extract and convert students' identity numbers and handwritten marks on assessment forms into a spreadsheet. The training data contain three sets of LBP values for each digit. The recognition rate of handwritten digits using LBP is about 50% because LBP could not fully describe the structure of the digits. Instead, LBP is useful in term of scaling the digits `0 to 9' from the highest to the lowest similarity score as compared with the sample using chi square distance. The recognition rate can be greatly improved to about 95% by verifying the ranking of chi square distance with the salient structural features of digits.