Detection of Orientation-Modulation Embedded Data in Color Printed Natural Images

Abstract

This article addresses methods for detection of orientation-modulation data embedded in color dispersed-dot-halftone images. Several state-of-the-art methods for detection of orientation-embedded data in printed halftone images have been proposed, however they have only been evaluated independently without comparing with each other. We propose an improved detection method, which is using Principal Component Analysis (PCA) components as oriented-feature extractors, and a probabilistic model for the print-and-scan channel for maximum likelihood detection. The proposed detector and four state-of-the-art detectors are compared with each other in terms of correct detection rate, using a comprehensive testing set of printed natural images captured with three different devices. The proposed detector achieves highest correct detection rate using fewer feature extractors than the other methods, and it is significantly more robust to non-calibrated devices used for capturing the printed images. This is mostly due to the improved PCA-based oriented-feature extractors that are responsive to the embedded orientations and robust and insensitive to the other visual content.