Document image analysis with cooperative interaction between layout analysis and logical structure analysis

When a printed document is to be input to a computer system, the document must be converted to a computer-readable format, e.g., ASCII, PDF, RTF, CSV, or SGML/XML/HTML-tagged data. In order to obtain these data formats from a printed document, it is necessary to extract from the printed document as much information as possible, i.e., layout structure (layout objects and their hierarchical relationship), logical structure (logical objects and their reading order), and contents (text as OCR results). Logical structure analysis is concerned with attaching logical meanings to the layout structure and determining the reading order. The logical structure may be extracted hierarchically from the layout structure that is extracted by the layout analysis process from a document image. In this hierarchical analysis, some ambiguities cannot be reduced in the result obtained by a single functional process due to insufficient information. For example, sparse, irregular, or unconfined layout styles such as unpredictable non-text elements, program lists, mathematical expressions, or tables without ruled lines might not be analyzed correctly using only geometric information in the layout analysis process. If processes are connected sequentially in an entire system, system performance deteriorates due to accumulation of the errors of all the processes. In this paper, the author proposes a new document image analysis method with cooperative interaction between layout analysis and logical structure analysis to resolve the above problems. The proposed method has the following three advantages compared with other methods. (1) It can extract logical structures from imperfect layout structures due to large variation in formats, noise, or erroneous feature detection. Erroneous layout structure is automatically detected and is resolved by cooperation between layout analysis and logical structure analysis. (2) It breaks through previous limitations in layout analysis capabilities, because high-level logical information can be used in layout analysis. After this layout analysis, the global column structure of a document is extracted from the logical structure. (3) It can extract the logical structure of a document accurately, because the information to be used for logical structure analysis is obtained from the orderly column structure and is made more accurate by the cooperative interaction between layout analysis and logical structure analysis.