Digital forensics of visual-based evidence from video surveillance systems and forensic photographs holds object detection as a key aspect of the process. Recognizing an instance of object classes over a wide range of image data using computational techniques is one of the areas that has gained continuous attention over the years due to their numerous practical applications. Several algorithms and techniques have been specified for object detection and recognition with Machine Learning gaining more prominence and ensuring the remarkable performance of object detection and recognition systems. This study presents a comprehensive review of the frameworks and applications of Machine Learning in object detection and classification with particular applications to Digital Forensics. The analysis covers a wide range of publications between 2007 and 2019 available in different indexed and non-indexed databases and the candidate papers were selected using certain exclusion criteria proposed in the Kitchenham’s methodology. The study in a bid to streamline future researches categorized digital forensic researches into six knowledge areas and identified the convolutional neural network as a state-of-the-art algorithm for machine learning-based digital forensics.

The State of the Art in Machine Learning-Based Digital Forensics

Digital Video authentication is very important issue in day to day life. A lot of devices have got the ability of recording or capturing digital videos and all these videos can be passed through the internet as well as many other non-secure channels. There is a problem of illegal updating or manipulation of digital video because of the development in video editing software. Therefore, video authentication techniques are required in order to ensure trustworthiness of the video. There are many techniques used to prevent this issue like Digital Signature and Watermarking, these solutions are successfully included in copyright purposes but it’s still really difficult to implement in many other situations especially in video surveillance. In this paper, a new method called PLEXUS is proposed for digital video authentication on temporal attacks. In authentication process, the sender will generate a signature according to the method steps using a video and private key. In verification process, the receiver will also generate a signature using the same video and private key then each signature will be compared. If the two signatures are matched then the video is not tampered otherwise the video is tampered. This method is implemented using 10 different videos and proved to be an efficient method.

https://thesai.org/Downloads/Volume9No11/Paper_104-Video_Authentication_using_PLEXUS_Method.pdf

Video Authentication using PLEXUS Method

The scheme developed the method to protect the copyright of digital video while encoding the video through MPEG standard. The technique suggested to insert the same watermark at two different locations in intra-coded frame. As these frames have highest degree of luminance part as comparision to Bi-directional(B-frames) and Predicting-frames (P-frames), therefore a little alteration to such frames doesn’t degraded the perceptibility of proposed video watermarking method. The scheme also tested the robustness by applying intentional and unintenional attacks which provides satisfactory results for most of the attack A reasonable size of watermark bits can also be adjusted in a single I-frames as it contains the sufficient number of quantized discrete cosine transform block therefore great size of watermark can be accomodated in the video object.

Securing Copyright of Digital Video by Exploiting Quantized DC coefficients

The immense quantities of information is uploaded and downloaded on the internet causing society to accept everything it sees as true without questioning its integrity, security and privacy, and authentication. There are many different methods and techniques, which are suggested for overcoming these problems. it will help a lot to restrict the modified use of the content developed by the creator. With this, the creator can track anybody else's claim for the content, illegal distribution of the content. This paper throws light on some techniques used for video authentication by several authors.

A Review on Video Authentication Technique Exploiting Watermarking Methods

Determining video authenticity has become a challenging task in video forgery detection and localization systems. One of the core difficulties is to extract a suitable feature descriptor. The recent studies on passive object-based video authentication are limited to localize the forged frames in copy-paste/delete object forgery. That is done without considering the object motion interpolation. It performs video authentication on simple videos with a static camera without using dynamic texture. This research work proposes an Action Passive Object-based Video Authentication Algorithm (APOVA). The APOVA consists a new statistical texture descriptor and a forged keyframes localization algorithm to validate video data. The APOVA algorithm has been tested on the Shih-Tang dataset. The experimental results shows that APOVA can obtain the best results in accuracy using the proposed descriptor. In addition, it is able to achieve the best performance in forged keyframes localization with F 0.5 _score = 0.93 and video authentication with F 0.5 _score = 0.99. As a result, APOVA based on APOKL is considered suitable algorithm for complex color video authentication, especially for passive object-based video forgery systems.

Passive Object-based Video Authentication Using Stereo Statistical Descriptor on Wavelet Decomposition

With the outgrowth of video editing tools, video information trustworthiness becomes a hypersensitive field. Today many devices have the capability of capturing digital videos such as CCTV, digital cameras and mobile phones and these videos may transmitted over the Internet or any other non secure channel. As digital video can be used to as supporting evidence, it has to be protected against manipulation or tampering. As most video authentication techniques are based on watermarking and digital signatures, these techniques are effectively used in copyright purposes but difficult to implement in other cases such as video surveillance or in videos captured by consumer's cameras. In this paper we propose an intelligent technique for video authentication which uses the video local information which makes it useful for real world applications. The proposed algorithm relies on the video's statistical local information which was applied on a dataset of videos captured by a range of consumer video cameras. The results show that the proposed algorithm has potential to be a reliable intelligent technique in digital video authentication without the need to use for SVM classifier which makes it faster and less computationally expensive in comparing with other intelligent techniques.

M. Al-Athamneh

Papers

Video authentication based on statistical local information