Extraction and Identification of Bar Graph Components by Automatic Epsilon Estimation
Sarunya Kanjanawattana, Masaomi Kimura · International Journal of Computer Theory and Engineering · 2017
Clustering is an unsupervised learning technique primarily used to analyze data.Density-based spatial clustering of applications with noise (DBSCAN) is effective for image clustering because it clusters neighbor objects that are located within a radius of an Epsilon parameter.However, identifying this parameter correctly requires expert knowledge.We propose methods to estimate Epsilon values effectively based on the density of each area wherein objects are located in order to extract graph components, such as axis descriptions (e.g., Xand Y-axis titles) and legends.We verified axis description extraction by measuring accuracy, precision, recall, and F-measure.The results indicate that the proposed automatic Epsilon estimation method is reliable.To evaluate legend extraction, we compared the proposed automatic Epsilon estimation to a method using a default Epsilon value (i.e., 0.6).The results demonstrate that the proposed method returns suitable Epsilon values.The proposed parameter estimation method is capable of handling graph component extraction effectively.