PageSegMode in Tesseract

最新推荐文章于 2025-05-16 16:33:30 发布

windtalkersm

最新推荐文章于 2025-05-16 16:33:30 发布

阅读量2.5k

点赞数 1

分类专栏： Tesseract备忘 OCR备忘算法备忘

本文链接：https://blog.youkuaiyun.com/windtalkersm/article/details/72455758

版权

Tesseract的PageSegMode非常复杂，但一旦理解了其术语的含义，比如PSM_COL_FIND_ENABLED，就会变得清晰。在某些模式下，如PSM_SINGLE_COLUMN，会先尝试找到至少一列，否则后续步骤会失败。其他模式则直接进入识别阶段。特别地，PSM_SINGLE_COLUMN模式虽然也会调用某些检测，但其行为与其它模式不同，可能会使用不同的处理流程。而PSM_LINE_FIND_ENABLED宏对应的是单块垂直或水平对齐的文本处理。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

The PageSegMode in Tesseract is REALLY, REALLY, REALLY confusing!!

Well, it is kind of all make sense in the end, once one understand the exact meaning of the terminologies used in Tesseract Documentation, such as column, block, line…

The so-called Automatic page segmentation is supposedly verified by PSM_COL_FIND_ENABLED macro, for 1,2,3 mode

inline bool PSM_COL_FIND_ENABLED(int pageseg_mode) { return pageseg_mode >= PSM_AUTO_OSD && pageseg_mode <= PSM_AUTO; }

which will incur Tesseract::AutoPageSeg then ColumnFinder::FindBlocks and most importantly TabFind::FindTabVectors , which in theory should find at least one column.

Otherwise ColumnFinder::MakeColumns later