Table correction method and device based on spatial position relationship and multi-modal verification

Abstract

The embodiment of the specification provides a table correction method and device based on spatial position relationship and multi-modal verification, wherein the method fundamentally changes the traditional paradigm of relying on fixed rules or pure text matching by constructing an intelligent correction process dominated by spatial position relationship and integrating multi-modal verification. The method realizes the universal understanding and adaptive matching of various table structures by automatically extracting reference objects and analyzing the spatial topological relationship between fields, significantly improving the accuracy and robustness of field misplacement correction. The introduced multi-dimensional confidence evaluation system and hierarchical decision mechanism ensure that each correction operation has reliable basis, realizes the safety and controllability of automatic processing, and effectively eliminates the risk of "more mistakes after correction". Finally, the method integrates data integrity repair while completing error correction, outputs high-quality structured data, and greatly improves the end-to-end automation level and result reliability of the table recognition post-processing link.