Modern Approaches to Chemical Image Recognition

Abstract

Millions of existing patent documents and journal articles dealing with chemistry describe chemical structures by way of structure images (so-called Kekulé structures). While being human-readable, these structure images cannot be interpreted by a computer and are unusable in the context of most chemoinformatics applications: structure and substructure searches, chemo-biological property calculations, etc. There are currently many formats available for storing structural information in a computer-readable format, but the conversion of millions of images by hand is a cumbersome and time-consuming process. Therefore there is a need for an automatic tool for converting images into structures. One of the first such tools was presented at ICDAR in 1993 (OROCS). We would like to present modern developments in optical structure recognition which build upon the ideas developed earlier and add modern enhancements to the process of automatic extraction of structure images from the surrounding text and graphics and conversion of the extracted images into a molecular format. We describe in detail two top performing chemical OCR applications—one open source and one academic software package. The performance here was judged by TREC-CHEM 2011 and CLEF 2012 challenges.