A mahout based image classification framework for very large dataset

In this paper, we present a distributed computing framework for image classification towards the current challenge of image big data due to enormous streaming image data sources, such as image sharing over online social network and massive video surveillance streams from ubiquitous cameras all over our daily life. The proposed framework consists of four modules aiming at feature extraction, dimension reduction, bag of feature modeling, and supervised learning respectively. This distributed computing framework is implemented on Hadoop with Mahout support. We apply the framework for classifying whether a person is on calling or not in a surveillance video to verify the correctness and scalability.