Abstract: Skew document images often appears when it is captured by image acquisition devices such as cameras or scanners, which may induce recognition mistakes by Optical Character Recognition (OCR) software. The paper proposed an optimized method for the skew detection of document images, and its objective function is the image projection grating slit width. The document image angle is the inverse of the projection angle, when the corresponding projection grating slit width is the largest. The detection range is expanded and the detection speed is increased by the grating line width function. The amount of calculation in detection is decreased by preliminary projection on equispaced rows and back projection. The detection precision is improved by dichotomy. In the experiments document images where a few illustrations were used, and the skew detection results show the proposed method is of high efficiency and robustness.

Key words: document image, skew detection, projection, grating slit, optimization

摘要： 针对由照相机扫描仪等文档获取设备拍摄的文档图像可能存在倾斜，进而导致光学字符识别（OCR）软件不能正确识别的情况，采用了一种以文档图像投影栅缝宽为目标函数的优化方法，栅缝宽最大值对应的投影角度的相反数即为文档图像的倾斜角。利用栅线宽函数扩大了检测范围，并提高了检测速度；利用反投影法和均布列预投影等方法，减少了计算量；利用二分法提高了算法的检测精度。通过一些包含少量插图的文档图像的倾斜角检测实验，验证了该方法的有效性。

关键词: 文档图像, 倾斜检测, 投影, 栅缝, 优化