Persian Optical Character Recognition Using Deep Bidirectional Long Short-Term Memory

Optical Character Recognition (OCR) is a system of converting images, including text,into editable text and applied to various languages such as English, Arabic, Persian. While these have similarities, their fundamental differences can create unique challenges. In Persian, continuity between Characters, the existence semicircles, dots, oblique, left-to-right characters English words in context are some most important challenges designing Persian OCR systems. Our proposed framework, Bina, designed special way address issue by utilizing Convolution Neural Network (CNN) deep bidirectional Long-Short Term Memory (BLSTM), type LSTM networks that has access both past future context. A huge diverse dataset, about 2M samples contexts,consisting fonts sizes, also generated train test performance model. Various configurations tested find optimal structure CNN BLSTM. The results show Bina successfully outperformed state art baseline algorithm achieving 96% accuracy 88% contexts.