Semantic coherency: the basis of an image interpretation device-application to the cadastral map interpretation

Abstract

This paper describes the formal architecture of a device capable of interpreting technical and cartographic documents. Our approach is two-fold, based on a model of the document and on the implementation of a set of "builders", the aim of which is to progressively construct information of as high a semantic level as that provided by the document drawer. The implementation support is the French cadaster. The interpretation process is performed in two stages: the first one consists in constructing the information, through a bottom-up approach. Then, this information is analyzed and the set of objects considered as nonvalid ("inconsistent") with regard to the document model are reexamined through a cycling stage. The whole approach is presented, and the first implementation has enabled us to quantify the interpretation results and to verify the relevance of the cycling stage.

Capturing Raster Data From Scanned Thematic Maps Using Desktop Graphics Software

“…Some work has been done to develop automated data capture techniques using a scanned image of the map. However the complexity of most maps means that current image recognition methods can only deal with certain, simple map types (Ogier et al 1998, 2000 or with only one map element, such as text or linework (Yan and Wu 1994, Mariani et al 1995, Chen 1999, Gamba and Mecocci 1999. Full automation still seems a distant goal, and so some authors have explored semi-automated approaches, in which a human operator still plays a key role but is provided with software tools to automate parts of the capture process.…”

Shall deep learning be the mandatory future of document analysis problems?

“…Here Marr and Gestalt theory can be cited [68]. This was the foundation of original studies such as Ogier in 2000 [69]. Ogier proposed a complete cadastral map interpretation system based on a perception cycle principle.…”

Knowledge-driven understanding of images in comic books

“…It is not easy to combine different related sub-tasks. Previous work concerned real scene image analysis [5], retrieval [8] and understanding [30,12], medical image annotation using description logic and inference engine [22], object-based image retrieval [36,48] (between keyword-based and query-by-example) and image interpretation [23,37]. Most researchers agree on the importance of using the underlying semantic of topological information, spatial distances, directional relative position and complex relations such as "between", "surround" and "among", to obtain an exhaustive description list of the contents for a given context.…”