Image classification using EXIF metadata

Abstract

Semantic classification of unrestricted images is still an open problem regardless of all efforts done. Present methods have only mainly focused on features extracted from the image content (e.g. colour, texture, shape). Conversely, EXIF metadata recorded by the camera can be exploited to aid the classification process. Demonstrating scenery-object classification as an example, analysis of results has revealed different combinations of metadata features that contribute variedly in predicting scenery-object image classification when using different classifiers. The evaluation was done by using machine learning to which a dataset of 500 digital images, consisting of 250 random scenery images and 250 random object based images were trained and tested. For classification, Support Vector Machine (SVM) and K-Nearest Neighbour (KNN) classifiers were used. The research project achieved a result as high as 75.15% for the classification of scenery-object image using KNN.