Dissimilarity Based Regularized Deep Learning Model for Information Charts

The charts are very much convenient way to represent the complex data into simple pictorial based representation. Every chart type has variations in its characteristics, structure, and appearances making every type and subtype of chart different from each other on its physical outlook. Classification of such similar outlook charts still remains an untouched area. This paper presents a model that computes chart dissimilarity index, which is amalgamated with regularization on input layers of the learning model. Thus, all structural variations of charts are integrated into the model which produces 96.66% accuracy rate outperforming existing state-of-the-art models. We proposed a novel approach to learn structure invariant dissimilar features using regularized learning techniques, by incorporating the dissimilarity index with the learning model to ease in learning dissimilar and hidden features of chart images.