Digital Image Forensics Lecture 4

Study Notes

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).

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.

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.

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.

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

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.

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.

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 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).