Research on Improving OCR Recognition Based on Bending Correction

With the emergence of smart phones, the traditional document image capture method has gradually evolved into the handheld camera shooting. After the shooting, a series of problems such as perspective distortion, uneven illumination and lens distortion etc. have brought many technical challenges to the recognition of document image in OCR. On page curl result in the text of the distortion, this paper proposes a proposed a document image to distort algorithm based on continuous skeleton image, first collected text image filtering, and morphological processing, it is concluded that the text of the minimum circumscribed rectangle to calculate the slope of the text, then further morphological operation to calculate the text used the baseline, combining three times bezier curve correction, complete text line finally uses tesseract’s own library to the text of the corrected image recognition, get the data information of the text. Through experiments, the modified algorithm has a better correction effect on character distortion, and it takes time to correct a single frame image, which can improve the recognition rate of OCR and meet the needs of daily life.