Reverse Data Interpretation Of Visual Charts

Visual charts are often used as a graphical representation to facilitate the understanding of digital data.However,for some ancient charts,the original data may have been lost.In addition,in the scientific literature,the raw data of most charts are also not available.These problems hinder the design of more effective visualizations and the process of further analysis of the chart.This paper provides an effective analysis method for the reverse data interpretation of charts.This paper implements Chart Decoder,a system that implements visual feature decoding and recovers data from chart images.By using the visual chart image as input,the system applies deep learning techniques to complete chart type classification.Then by using computer vision and text recognition technology,the system complete automated data extraction from chart images.Chart Decoder can identify five types of charts(bar chart,pie chart,line chart,scatter chart and radar chart)using deep learning-based classifiers with a classification accuracy of over 99%.Additionally,for text information extraction,this paper provides an effective text localization and recognition method,and implements text role classification.In order to evaluate the effectiveness of the proposed algorithm,the paper evaluates the system on two corpora: 1)visual charts collected from the network,and 2)chart images automatically generated by a program script.The results demonstrate that the system is able to recover data from bar charts with a high rate of accuracy.