Ensemble of Machine Learning Classifiers for Improved Image Category Prediction Using Fractional Coefficients of Hartley and Sine Transforms

Social networking sites have given rise to tremendous volume of images, which implies need of proper organisation of image databases with efficient retrival and categorisation mechanisms. Images if stored in appropriate fashion may help to reteive them fastly and correctly as and when required. Image category prediction with the proposed machine learning based approach palys an important role in visual content based image category prediction for efficient handling of voluminous data. The system uses the content as transformed fractional coeficients to generate the feature vector for image class peridication using Sine and Hartley transforms. Machine learning algoritms alias Random Forest, SVM, Simple logicts are employed for proposed image category prediction method. The paper also proposes ensembling of these machine learning alogorithms with majority voting at decision level for improved image category prediction. The experimentation is carried out on the fraction of the standard image dataset. The result analysis show that the fractional transform coefficients gives the capability for better discrimination than that of consideration of all transformed coefficients considered to form feature vector; as indicated by higher image category prediction accuracy values. Also it has been observed that ensembling of machine learning algorithms (Random Forest, Simple Logistic and SVM) has given best classification accuracy of 72.91% with Hartley transformed fractional content as features.