A Component-detection-based Approach for Interpreting Off-line Handwritten Chemical Cyclic Compound Structures

Recently, off-line handwritten chemical structure formulas recognition draws considerable attention from re-searchers for its research challenges and supports to numerous in-teresting applications in education. The existing solutions belong to the rule-based category or the end-to-end trainable category, which suffer from the problems of low generalization capability or lacking the accurate alignment information between the input and output. To tackle these problems to enable auto correction of handwritten chemical assignments at a fine-grained level, in this paper we propose a component-detection-based approach for interpreting the spatial structure of off-line handwritten chem-ical cyclic compound structure formulas. Specifically, we define different components of cyclic compound structure formulas as objects (including graphical objects and text objects) and adopt the deep learning detector to detect them. Then, considering the property of chemical notations, we propose the Non-Maximum Area Suppression (NMAS) algorithm to improve the detection results. Finally, with these detection results and the geometric relationships of detected objects, this paper design a holistic algorithm for interpreting the spatial structure of handwritten cyclic compound structure formulas. The proposed method is evaluated on a self-collected dataset consisting 2100 samples and achieved an accuracy of 89.5%