Proposes an end-to-end architecture replacing standard CNN backbones with Swin Transformer to capture global image context. Introduces Multi-label Focal Loss to handle severe token imbalance in chemical datasets.
This paper reviews three decades of OCSR development, transitioning from rule-based heuristics to early deep learning approaches. It includes a benchmark study comparing the performance of three open-source tools (OSRA, Imago, MolVec) on four diverse datasets.
Describes chemoCR, a system that converts bitmap chemical diagrams into connection tables using a pipeline of texture-based vectorization, OCR, and a rule-based expert system, achieving 65.6% perfect recall on the TREC 2011 task.
ChemReader achieved 93% accuracy on the TREC 2011 Image-to-Structure task, with detailed error analysis revealing the need for improved chemical intelligence in bond recognition and node merging algorithms.
A resource paper detailing the CLEF-IP 2012 benchmarking lab. It introduces specific IR tasks for patent processing along with ground-truth datasets.
Describes the MolRec system’s performance in the CLEF 2012 Chemical Structure Recognition task, detailing its rule-based vectorization engine and analyzing failure modes like touching characters and complex bond types.
Benchmarks OSRA on CLEF-IP 2012 patent data, showing native image processing improves precision from 0.433 to 0.708 over external splitting tools. Describes OSRA’s pairwise distance algorithm for segmentation that handles overlapping molecules better than bounding boxes.
This resource paper details the third TREC Chemical IR campaign, introducing a novel Image-to-Structure task and analyzing 36 runs from 9 groups to benchmark chemical information retrieval.
ChemInk introduces a sketch recognition system for chemical diagrams that combines multi-level visual features via a joint Conditional Random Field (CRF), achieving 97.4% accuracy and outperforming CAD tools in user speed.
This paper introduces CLiDE Pro, an advanced OCSR system that segments document images and reconstructs chemical connection tables. It features novel handling for crossing bonds and generic structures, validating performance on a publicly released benchmark of 454 scanned images.
This paper details the algorithmic pipeline of OSRA, an open-source tool that converts raster images of chemical diagrams into connection tables (SMILES/SDF). It outlines specific heuristics for page segmentation, vectorization, and atom recognition used in the TREC-CHEM Image2Structure task.
This methodological paper presents a system for digitizing chemical images into SDF files. It utilizes a custom vectorization algorithm and chemical rule validation, achieving 94% accuracy on benchmark datasets compared to 50% for commercial tools.