Sensor Pattern Noise Estimation Based on Improved Locally Adaptive DCT Filtering and Weighted Averaging for Source Camera Identification and Verification
Photo Response Non-Uniformity (PRNU) noise is a sensor pattern noise characterizing the imaging device. It has been broadly used in the literature for source camera identification and image authentication. The abundant information that the sensor pattern noise carries in terms of the frequency content makes it unique, and hence suitable for identifying the source camera and detecting image forgeries. However, the PRNU extraction process is inevitably faced with the presence of image-dependent information as well as other non-unique noise components. To reduce such undesirable effects, researchers have developed a number of techniques in different stages of the process, i.e., the filtering stage, the estimation stage, and the post estimation stage. In this paper, we present a new PRNU-based source camera identification and verification system and propose enhancements in different stages. First, an improved version of the Locally Adaptive Discrete Cosine Transform (LADCT) filter is proposed in the filtering stage. In the estimation stage, a new Weighted Averaging (WA) technique is presented. The post estimation stage consists of concatenating the PRNUs estimated from color planes in order to exploit the presence of physical PRNU components in different channels. Experimental results on two image datasets acquired by various camera devices have shown a significant gain obtained with the proposed enhancements in each stage as well as the superiority of the overall system over related state-of-the-art systems.
PROJECT OUTPUT VIDEO:
In the literature, there has been a growing body of research devoted to source camera identification using the PRNU. The PRNU estimation process can be divided into three stages, i.e., the filtering stage, the estimation stage, and finally the post estimation (enhancement) stage. In the filtering stage, a pattern residual signal, also called the noise residue is obtained from each image through the difference between the input image and its filtered version.
The most widely known system was initially developed by Lukas et al. in order to identify the origin of digital images using the PRNU. It uses a set of images to extract a noise residue from each image. The estimated noise residues are then averaged to obtain a camera reference PRNU noise.
In other work, the Maximum Likelihood Estimator (MLE ) was applied to estimate the camera reference PRNU.
In other existing work, the authors proposed preprocessing steps to enhance the commonly used PRNU through Wiener filtering and zero-mean operations.
DISADVANTAGES OF EXISTING SYSTEM:
The rational is that there are artefacts which may be shared by different cameras of the same model or brand and this leads to a rise in false identification rates
This work addresses the problem of source camera identification and verification in image forensics based on PRNU estimation. It is worth mentioning that the PRNU is the result of imperfections caused by the manufacturing process due to the lack of homogeneity of the silicon area in the imaging sensor.
The noise due to sensor imperfections is a weak signal of the same size as the output image. First, digital images are considered in the form of separate color channels. Then, an improved version of the LADCT de-noising filter is applied to reduce the effect of scene details on noise residues. Next, for efficient sensor pattern noise estimation, the obtained noise residues are averaged using the proposed WA technique.
Finally, we propose to concatenate the PRNUs estimated from the primary color planes in order to exploit the presence of physical PRNU components in different color channels.
ADVANTAGES OF PROPOSED SYSTEM:
In camera identification, the noise residue of a query image is compared to all PRNUs stored in the database. The closest PRNU corresponds to the camera which has been used to take the image.
In camera verification, however, the similarity between the noise residue and the PRNU of a certain camera is compared to a given threshold in order to verify whether the image is originated by the camera.
An experimental analysis covering application scenarios in digital image forensics has shown the superiority of the proposed system over state-of-the-art techniques.
- Image Acquisition
- Image denoising
- Extraction Noise Residue and PRNU
- Identification and Verification
Images are acquired from Gallery.
In this process, RGB color image splited into 3 Channel. R channel, G channel and B channel was extracted.
After preprocessing stage, separated channel are used for image denoising process. DCT is used for denosing process. First add noise in each channel. Then improved locally DCT is applied R, G and B channel. Finally get denosied R Channel, G Channel and B Channel.
Extraction of Noise Residue and PRNU:
After image denoising process, extraction of PRNU is performed.
First R channel image is divided into 8*8 blocks. Then find mean and noise variance for each blocks.
Second G channel image is divided into 8*8 blocks. Then find mean and noise variance for each blocks.
Third B channel image is divided into 8*8 blocks. Then find mean and noise variance for each blocks.
Finally weighted average is generated for noise varaince of each chennel. This is called as PRNU.
Identification and Verification:
In this stage, camera identification and verification is performed based on PRNU. Distance function is used for identification process. Distance is calculated between Test PRNU and Database PRNU. Finally camera id is identified.
System : Pentium Dual Core.
Hard Disk : 120 GB.
Monitor : 15’’ LED
Input Devices : Keyboard, Mouse
Ram : 1GB.
Operating system : Windows 7.
Coding Language : MATLAB
Tool : MATLAB R2013A
Ashref Lawgaly, and Fouad Khelifi, Member, IEEE, “Sensor Pattern Noise Estimation Based on Improved Locally Adaptive DCT Filtering and Weighted Averaging for Source Camera Identification and Verification”, IEEE Transactions on Information Forensics and Security, 2016