Showing 1–2 of 2 results for author: Wong, Y J

Local Statistics for Generative Image Detection

Authors: Yung Jer Wong, Teck Khim Ng

Abstract: Diffusion models (DMs) are generative models that learn to synthesize images from Gaussian noise. DMs can be trained to do a variety of tasks such as image generation and image super-resolution. Researchers have made significant improvement in the capability of synthesizing photorealistic images in the past few years. These successes also hasten the need to address the potential misuse of synthesi… ▽ More Diffusion models (DMs) are generative models that learn to synthesize images from Gaussian noise. DMs can be trained to do a variety of tasks such as image generation and image super-resolution. Researchers have made significant improvement in the capability of synthesizing photorealistic images in the past few years. These successes also hasten the need to address the potential misuse of synthesized images. In this paper, we highlight the effectiveness of computing local statistics, as opposed to global statistics, in distinguishing digital camera images from DM-generated images. We hypothesized that local statistics should be used to address the spatial non-stationarity problem in images. We show that our approach produced promising results and it is also robust to various perturbations such as image resizing and JPEG compression. △ Less

Submitted 25 October, 2023; originally announced October 2023.

Numerical Evidence of Exponential Mixing by Alternating Shear Flows

Authors: Li-Tien Cheng, Frederick Rajasekaran, Kin Yau James Wong, Andrej Zlatos

Abstract: We performed a numerical study of the efficiency of mixing by alternating horizontal and vertical shear ``wedge'' flows on the two-dimensional torus. Our results suggest that except in cases where each individual flow is applied for only a short time, these flows produce exponentially fast mixing. The observed mixing rates are higher when the individual flow times are shorter (but not too short),… ▽ More We performed a numerical study of the efficiency of mixing by alternating horizontal and vertical shear ``wedge'' flows on the two-dimensional torus. Our results suggest that except in cases where each individual flow is applied for only a short time, these flows produce exponentially fast mixing. The observed mixing rates are higher when the individual flow times are shorter (but not too short), and randomizing either the flow times or phase shifts of the flows does not appear to enhance mixing (again when the flow times are not too short). In fact, the latter surprisingly seems to inhibit it slightly. △ Less

Submitted 29 October, 2021; originally announced November 2021.