Smart Cross-Platform Binary Visualisation Tool
Bee Lee Ong, Chai Kiat Yeo · 2018
Binary visualisation (binvis) is superseding the traditional way of file analysis. Currently, there are efforts to develop some intelligence to detect certain binary patterns in arbitrary files, which mostly utilise simple classifiers on raw byte sequences. However, none of them explored the use of intelligence to process the visuals generated from the binvis tools. Hence, this paper details the proof-of-concept (POC) of a smart, cross-platform binvis tool, written in Java that uses an open source deep learning library. In addition, a web crawler was created to obtain a large sample of the common file types used on the internet. These files were processed into digraph visuals which were fed into a convolutional neural network implemented via the library. However, the classification labels used and the existence of compressed data (which added noise to the resulting digraph visuals) meant that the classification was not very accurate. Nevertheless, the POC shows that deep learning for binvis could be further explored by using classification labels based on the fundamental data formats (e.g. text, bitmap image, etc.) and supplementing the digraph data with some other known visual techniques.