Traffic classification based on visualization

Nowadays application based on encryption flows are increasing. Such applications are beneficial to protect the privacy but also offer convenience to hackers to avoid detection. This paper discusses the feasibility of applying a visualization technique to recognize traffic flows without reading payloads. We proposed a method to extract features from packet size and intervals and then change them to a 2-D image. Unlike most of machine learning methods which use features directly, we enhance the images with a modified mountain function to explore the potential of flow features. Finally principle component analysis is used to classify the traffic flows based on pattern recognition. The result shows that the images generated from flows can be recognized more easily after image enhancement.