Video-Based Evidence Analysis and Extraction in Digital Forensic Investigation

In forensic investigation, digital cameras and mobile devices are routinely seized as evidence sources. Video and images retrieved from these devices are widely used in crime evidence investigation, which can provide key forensic evidence items, piece together existing evidence items, or establish links between evidence items in particular case. The Closed circuit television (CCTV) systems are widely used for malls, banks, traffic intersections to park, stores, or even home, where video evidences are retrieved from these systems can be used as evidences much more than ever before [1]. Along with the use of smart devices, such as mobile phone, smart watch, fitkit, etc., audio and video evidences can be easily available in investigation.

In the past few years, the ‘image enhancement’ techniques have been proposed [2]–[6], most of them can be grouped into spatial domain methods and frequency domain methods.
These techniques shows good potential to improve the quality of images, but only a few of them can be used for low quality of footage, such as cctv footage, mobile video clips, etc. [7], [8]. Many cctv surveillance systems export footage in their own formats, which need to be re-format or converted to a suitable format that easier for investigation. However, this can often cause the lower of quality and information loss, which makes the examination process difficult. The footage in digital forensics is often used for comparative analysis, including forensic analysis, comparison of
images of questioned about know objects such as subjects, vehicles, clothing, and weapons, with expert opinion being providing on the findings [9]–[11]. In many modern CCTV systems, facial recognition services are embedded to identify online criminals or suspects [26].

Other services such as motion detection, body and face recognition, cross-pose recognition, gait recognition,
are widely researched in the past few years. In some hard cases (poor viewing conditions), it is very difficult to identify humans take advantage of face, body, still, etc. Although
many image processing techniques have been developed in the past few decades, most of them do not take advantage of face, body, etc. In video based forensic investigation, following challenges still need to be addressed

  1. Forensic identification, in digital forensics investigation, probe image often ‘‘different’’ from gallery images due to heterogeneous face recognition due
    to the low quality, angle of camera, color, etc., new techniques that to improve the quality of footage need to be developed;
  2.  Establish the links between objects in investigation scenarios and related available evidence resources, such as cctv footage, online image records, or the history trace of these objects;
  3.  Due to the emerging technologies, such as social networks, internet of things (IoT), mobile devices, etc., the new investigation techniques are not only ‘end all’ solutions for law enforcement but tools make use of whatever data is available;
  4. Intelligent techniques, such as deep learning, artificial intelligence, can assist the digital forensics to quickly identify potential evidences from existing sources collected;
  5. Robust evidence extraction methods, such as robust recognition, subject detection, etc. , are still need to be developed for forensic investigation, for example,
    in video-based facial recognition, to detect unfamiliar faces involves facial ageing, marks, forensic sketch recognition, and near-infrared face recognition.

It is clear that the digital forensics investigation over video related resource highly depends on the quality of footage recordings, poor quality would significantly reduce the confidence level of the investigation process and thus would not make a strong evidence to be presented in a court. In this work, we will address the above challenges and propose
new methods aim at assisting effective video-based digital forensics investigation.

The main contribution are three-fold:

1) An video-based digital investigation framework is proposed that addresses the concerns in low quality of  footage, establishment of links between objects and
available digital evidences, detection techniques over video footage, and intelligent techniques that can be used in modern digital forensics;
2) Detailed quality improvement method is provided for low quality footage, include the adaptive histogram equalization (AHE), contrast limited AHE (CLAHE), etc.; and a test scenario is provided to compare the different algorithms;
3) A deep learning based object identification scheme in footage is proposed that can be used to establish the links between the objects, subjects, and their behaviors
in the available footages.

The main aim of forensic video analysis is to identify strong evidence items at different level. In this paper, we focus on the contents of the video to develop efficient video analysis
techniques from the view point of forensics.

FORENSIC VIDEO ANALYSIS FRAMEWORK

As a result of the popularity of smart mobile devices and the low cost of surveillance systems, visual data are increasingly being used in digital forensic investigation. Digital videos have been widely used as key evidence sources in evidence identification, analysis, presentation, and report. The main goal of this paper is to develop advanced forensic video analysis techniques to assist the forensic investigation. We first propose a forensic video analysis framework that employs an efficient video/image enhancing algorithm for the low quality of footage analysis. An adaptive video enhancement algorithm based on contrast limited adaptive histogram equalization (CLAHE) is introduced to improve the closed-circuit television (CCTV) footage quality for the use of digital forensic investigation. To assist the video-based forensic analysis, a deep-learning-based object detection and tracking algorithm are proposed that can detect and identify potential suspects and tools from footages.