A Multi-view Convolutional Neural Network Approach for Image Data Classification

Multi-view learning promises to enhance the classical machine learning algorithm performance with the optimal setting. Currently, the Convolution Neural networks (CNN) is a widely utilized algorithm for image data for feature extraction and classification. The applicability of multi-view learning to enhance the CNN may be analyzed. Therefore, this research has focused to develop a novel approach, called Multi-view convolutional neural network (MvCNN) that enhances the classical CNN performance. The applicability has been analyzed by experiments over nine standard image datasets, where MvCNN performance has compared with single-view learning of CNN (SvCNN) successfully. The comparative analysis with classification accuracy and non-parametric statistical analysis shows that the performance of MvCNN is better than SvCNN.