Mishchenko, Ales; Vassilieva, Natalia
Keyword(s):Chart recognition; model-based classification; image analysis; comparative evaluation
Abstract: Recognition and classification of charts is an important part of analysis of scientific and financial documents. This paper presents a novel model-based method for classifying images of charts. Particularly designed chart edge models reflect typical shapes and spatial layouts of chart elements for different chart types. The classification process consists of two stages. First, chart location and size are predicted based on the analysis of color distribution in the input image. Second, a set of image edges is extracted and matched with chart edge models in order to find the best match. The proposed approach was extensively tested against the state-of-the-art supervised learning methods and showed high accuracy, comparable to those of the best supervised approaches. Advantages of the proposed model-based approach are that it doesn't require the supervised learning and that it uses the high-level features which are necessary for further steps of data extraction and semantic interpretation of chart images.
External Posting Date: October 6, 2011 [Abstract]. Approved for External Publication
Internal Posting Date: October 6, 2011 [Fulltext]