Interpretation and Segmentation of Chart Images Using h-Means Image Clustering Algorithm

For chart analysis, it is necessary to segment regions of interest from the background for a better understanding of distinctive pieces of charts. Many different methods, such as thresholding, clustering, compression-based methods, etc., are proposed for image segmentation. For accurate representation, decoding of graphical data or visual contents is needed for a readable and understandable data pattern for a computing machine. The existing works for classification of charts have relied on local and handcrafted features that lack in dealing with a massive amount of visual data. That could include substantial variabilities of complex data from within-class and between-classes of data. Therefore, we propose a segmentation method using histogram to segment the charts into a region of interest for feature extraction and classification. H-means algorithm uses histogram to decide the number of clusters by extracting peaks as ROIs for each chart image and segments foreground from the background. An image database of info charts having 5280 images covering ten categories of chart images is introduced to the research community.