Flowchart recognition for non-textual information retrieval in patent search

摘要

The paper discusses a very specific problem in information retrieval (IR) communities, namely, how to improve patent recognition by incorporating content such as drawings and sketches with particular emphasis on flowchart diagrams. The architecture and approach resembles the well-known content-based image retrieval (CBIR), with the intention of addressing CBIR's semantic gap, which occurs between queries between similar images. This is, at least, the impression given in the introduction. In actuality, the authors discuss an approach and architecture, which mostly focus on flowchart structural similarities rather than semantic ones. Furthermore, they mostly discuss how the two alternative branches, pixel-based (raster) and vectorial (vector graphics) representations compare with each other. The experiments and evaluation based on datasets from the CLEF-IP 2012 campaign does strengthen this viewpoint in that the flowchart recognition procedure is claimed to be improved.