Perspective Rectification For Camera-captured Document Images Based On Supporting Strokes

Currently, the technology of OCR for scan-oriented document images can efficiently realize the exchange from paper document to digital doucment. Along with the popularization of image acquisition equipment, such as digital camera, the camera-captured document image is becoming one of the main input sources of OCR system. Because the shooting angle is arbitrary and gathering environment is complex, camera-captured images often have a variety of distortion and make the recognition of OCR system ineffective. So it is necessary to carry on the correction of distortion before recognition.The rectification of camera-captured document images mainly includes the optical-distorted correction and the geometry-distorted correction. The perspective distortion is one of the most common geometric distortions. In view of the camera-captured documents image's perspective distortion, this article proposed one kind of correction method based on the support stroke. Firstly according to the characteristics of camera-captured document image's back-to-front noise interference and creased noise, the paper designs an algorithm of eliminating noise in the basis of analysis of stroke edge before perspective rectification. Through the k-means clustering and the stroke edge analysis, the algorithm can find out the noise location and deal with it. Then according to the optical principle and rectifying principle of the analysis of camera-captured document images'distortion, making use of apex of the quadrangle which is formed by Chinese character's supporting stroke and the horizontal border of text line as the optional controlling point, we can rectify the perspective document image by optimizing the option of controlling point. While the method also works for the rectification of skewing distortion.