Analysis of Machine Learning Based Scene Classification Algorithms and Quantitative Evaluation

摘要

Scene classification plays an important role in automated surveillance applications including pedestrian detection, indoor positioning, categorizing in semantic classes etc. The capability of machine learning to process big data with high speed make it suitable algorithm for supervised and unsupervised image/video processing applications. The scene classification processes large set of images to categorized it and hence weight-age is given to machine learning including in this paper. This paper provides indepth analysis of various machine learning based scene classification techniques including Sparse representation, Support Vectors Machine (SVM), Artificial Neural Networks (ANN), Convoluational Neural Networks (CNN), Spatial Pyramid Matching (SPM). The quantitative comparison of all algorithms with reference to type of datasets and corresponding accuracy are presented in the paper. The paper concludes that Caltech dataset has been used widely for performance testing and ResFeat algorithm achieves highest accuracy amongst all algorithms.