Extraction and analysis of features for identifying, clustering and modifying the source of images generated by mobile devices

Nowadays, digital images play an important role in our society. The presence of mobile devices with integrated cameras is growing at an unrelenting pace, resulting in the majority of digital images coming from this kind of device. Technological development not only facilitates the generation of these images, but also the malicious manipulation of them. Therefore, it is of interest to have tools that allow the device that has generated a certain digital image to be identified. The digital image source can be identified through the features that the generating device permeates it with during the creation process. In recent years most research on techniques for identifying the source has focused solely on traditional cameras. The forensic analysis techniques of digital images generated by mobile devices are therefore of particular importance since they have specific characteristics which allow for better results, and forensic techniques for digital images generated by another kind of device are often not valid. This thesis provides various contributions in two of the main research lines of forensic analysis, the field of identification techniques and the counter-forensics or attacks on these techniques. In the field of digital image source acquisition identification techniques, both closed and open scenarios are addressed. In closed scenarios, the images whose acquisition source are to be determined belong to a group of devices known a priori. Meanwhile, an open scenario is one in which the images under analysis belong to a set of devices that is not known a priori by the fo rensic analyst. In this case, the objective is not t he concrete image acquisition source identification, but their classification into groups whose images all belong to the same mobile device. The image clustering t echniques are of particular interest in real situations since in many cases the forensic analyst does not know a priori which devices have generated certain images. Firstly, techniques for identifying the device type (computer, scanner or digital camera of the mobile device) or class (make and model) of the image acquisition source in mobile devices are proposed, which are two relevant branches of forensic analysis of mobile device images. An approach based on different types of image features and Support Vector Machine as a classifier is presented. Secondly, a technique for the ident ification in open scenarios that consists of grouping digital images of mobile devices according to the acquisition source is developed, that is to say, a class-grouping of all input images is performed. The proposal is based on the combination of hierarchical grouping and flat grouping using the Sensor Pattern Noise. Lastly, in the area of att acks on forensic t echniques, topics related to the robustness of the image source identificat ion forensic techniques are addressed. For this, two new algorithms based on the sensor noise and the wavelet transform are designed, one for the destruction of t he image identity and another for its fo rgery. Results obtained by the two algorithms were compared with other tools designed for the same purpose. It is worth mentioning that the solution presented in this work requires less amount and complexity of input data than the tools to which it was compared. Finally, these identification t echniques have been included in a tool for the forensic analysis of digital images of mobile devices called Theia. Among the different branches of forensic analysis, Theia focuses mainly on the trustworthy identification of make and model of the mobile camera that generated a given image. All proposed algorithms have been implemented and integrated in Theia thus strengthening its functionality