Distortion Modeling and Invariant Extraction for Digital Image Print-and-Scan Process

After an image is printed-and-scanned, it is usually filtered, rotated, scaled, cropped, contrast-andluminance adjusted, as well as distorted by noises. This paper presents models for the print-and-scan process, considering both pixel value distortion and geometric distortion. We show properties of the discretized, rescanned image in both the spatial and frequency domains, then further analyze the changes in the Discrete Fourier Transform (DFT) coefficients. Based on these properties, we show several techniques for extracting invariants from the original and rescanned image, with potential applications in image watermarking and authentication. Preliminary experiments show the validity of the proposed model and the robustness of the invariants.