Print-Cam Image Distortion Correction for Robust Image Watermarking

At present, the widespread practice for image watermarking involves using handheld devices in unstructured conditions for image authentication. Print-cam watermarking specifically employs a smartphone camera to capture printed watermarked images. However, these captured images are typically not suitable for automatic information detection because of physical distortions, different camera angles, and varying illumination. Hence, rectifying distorted images is crucial to enhance the detection rate. In this paper, we presented Cam-Unet, an end-to-end neural network to directly predict the mapping from the distorted image to the rectified state. This learning framework is introduced to address print-cam issues specifically applied to identity images. Experimental results validate its superiority over state-of-the-art methods, with improved performance in terms of watermark detection rates. Tailored for industrial applications, Cam-Unet strikes an optimal balance between efficiency and cost-effectiveness, especially adept at addressing perspective distortions introduced by the print-cam process.