Line Chart Understanding with Convolutional Neural Network

Visual understanding of the implied knowledge in line charts is an important task affecting many downstream tasks information retrieval. Despite common use, clearly defining difficult because ambiguity, so most methods used research implicitly learn knowledge. When building a deep neural network, integrated approach hides properties individual subtasks, which can hinder finding optimal configurations for academia. In this paper, we propose problem definition explicitly chart and provide algorithm generating supervised data that are easy to share scale-up. To introduce data, set well-known modified convolutional networks evaluate their performance on real synthetic datasets qualitative quantitative analyses. results, extracted generated show patterns similar human-labeled data. This work expected separate scalable environment enhance into technical document understanding.