Digital Image Forensics Lecture 4

Digital Image Forensics

• Digital Image Forensics is a field that deals with the detection of tampered images.
• Common image manipulations include cloning (copying and pasting portions of an image to conceal objects or people) and resampling (resizing, rotating, or stretching portions of an image).

Pixel-based Techniques

• Cloning: two computationally efficient algorithms have been developed to detect cloned image regions.
• Resampling: a tampered image can be detected by analyzing the resampling artifacts introduced during the tampering process.

Format-based Techniques

• Double JPEG Compression: a manipulated image can be detected by analyzing the double compression artifacts introduced during the tampering process.
• JPEG Blocking: artifacts can be detected at the border of neighboring blocks in the form of horizontal and vertical edges.

Copy-Move Forgery Detection (CMFD)

• Common workflow consists of four stages: pre-processing, feature extraction, matching, and visualization.
• Pre-processing techniques include color conversions (e.g., RGB to grayscale) and block division (dividing the image into blocks).
• Feature extraction techniques include Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), and log polar transform.
• Matching stage seeks out similarities between two or more features in the image.
• Visualization stage displays and localizes the tampered regions in the forged image.

Pre-processing Techniques

• Color conversions: reduce dimensionality of the data and increase distinctive visual features in an image.
• Block division: reduces computational time for the matching process.

Feature Extraction

• Discrete Cosine Transform (DCT): a frequency transform technique that is robust to noise addition and JPEG compression.
• Discrete Wavelet Transform (DWT): a frequency transform technique that is robust to noise addition and JPEG compression.

Visualization Stage

• Block-based approach: visualizes the matching blocks by coloring or mapping the region of the matching blocks.
• Keypoint-based approach: displays the matching points by line transformation between each matching point.

Block-based Approach

• Splits an image into blocks for analysis during the pre-processing stage.
• Features are extracted from these blocks and compared against each other to determine the similarity between blocks within the image.

Frequency Transform

• Frequency transform is the most popular feature extraction technique for block-based approaches.
• Techniques include Discrete Cosine Transform (DCT), Fourier Transform, fast Walsh-Hadamard Transform (FWHT), Discrete Wavelet Transform (DWT), and Dyadic Wavelet Transform (DyWT).

Datasets

• UCID
• National Geographic
• ImageNet
• Kodak
• DOCR
• PIMPRCG
• USC-SIPI
• KSU
• Caltech-256