ITC-MNP: a diverse dataset for image file fragment classification

Image file fragment classification is a critical area of study in digital forensics. However, many publicly available datasets in this field are derived from a single source, often lacking consideration of the diversity in image settings and content. To demonstrate the effectiveness of a given methodology, it is essential to evaluate it using datasets that are sampled from varied data sources. Therefore, providing a sufficiently diverse dataset is crucial to enable a realistic assessment of any proposed method.

The dataset includes image file fragments of 4096 bytes from five formats (JPG, BMP, GIF, PNG, and TIFF), each processed with different conversion settings. The source images are categorized into three content types: Nature, People, and Medical. In total, the dataset contains 501,000 fragments. These fragments consist of file headers and incomplete end-of-file fragments, completed with random bytes to approximate how operating systems handle data when file sizes are not multiples of the sector size. This approach aims to simulate typical scenarios where fragments are recovered from a hard drive, though it may not capture all real-world complexities such as data corruption and complex file structures.