A study of binary local histogram correlation for automatic detection of optical printed Chinese character recognition errors

作者：Zhang, Jie; Xie, Ming^*; Luo, Daisheng

来源：Optik, 2015, 126(23): 4493-4500.

DOI：10.1016/j.ijleo.2015.08.002

• Optical printed Chinese character recognition
• Automatic error detection
• Differential low pass filter
• Image block division
• Error detection risk

In optical printed Chinese character recognition (OPCCR), some recognition errors would often occur, especially in recognition of similar characters. The recognition errors are commonly checked and corrected manually, because automatic detection of the errors is difficult. Hence, in this paper, we propose a binary local histogram correlation (BLHC) method for automatic detection of OPCCR errors. In the method, the recognized characters are converted into OPCCR images. The images of optical printed Chinese character (OPCC) and OPCCR are filtered using differential low pass filter (DLPF) and segmented into binary images using adaptive threshold segmentation. The binary images are divided into blocks and the histograms of the blocks are selected as the features of local structures. The correlations of the local histograms between OPCC and OPCCR are utilized to determine whether OPCCR errors occur. In image block division, three division methods are studied to see the affects on error detection risk (EDR) and error detection rate. To validate the BLHC method, some experiments have been done. Experiment results showed that BLHC method using 1/2 offset second division has the advantages of simple programming, fast computation and high error detection rate.

• 出版日期2015
• 单位四川大学