Classification of Binary File Fragments Using Texture Analysis

Abstract

Research up to date have been focused around the use of non texture based analysis approaches in an attempt to accurately classify the data type of le fragments. When les are viewed as a grayscale images their textures shows distinguishable characteristics from other data types. Here we employ texture analysis method known as gray level Co-occurrences matrix which have been successfully used in various texture analysis ap- plications to the problem of classifying data types of le fragments. We explored gray level quantization values from 4 to 64 with step increments of 4 and le fragment sizes of 8 8, 16 16, 32 32 and 64 64. All the 23 features that can be extracted from a GLCM matrix are not suitable for e cient and e ective data type discrimination. Therefore Sequential forward feature selection algorithm was used to determine the best feature combination for each gray level value and fragment dimension. The K nearest neighbor classi er was used as the classi er in this process. An impressive overall ac- curacy of 86.86% was achieved for the classi cation of 64 by 64 le fragments of 7 le types.