<p>Moiré pattern detection is an important issue on the reliability of visual data for maintaining private security. In this paper, inspired by the mechanism of moiré pattern formation, we find that if an image has the moiré pattern, the values of its RGB channels always appear as abnormal distribution patterns, which usually are observed as the significant imbalance. Based on this finding, we design a simple and effective feature extraction method, called the dominant RGB channel (DRC) coding, to capture the key characteristics of the images with moiré patterns. The coding results can be directly used to detect whether images have moiré patterns. Accordingly, we also design a specific multi-view deep convolutional neural network for moiré pattern detection (MCNN-MPD), a part of whose multi-view inputs are the DRC coding results. To improve the efficiency of using the training samples, we also incorporate the generative adversarial mechanism with MCNN-MPD to form a generative adversarial network for moiré pattern detection (GAN-MPD), whose generator is the original MCNN-MPD, and whose discriminator aims to distinguish whether its input is the pair of true image and its moiré pattern label. The numerical experiments support the effectiveness of the DRC coding method and the detection models that work with it, and show that these models outperform the existing learning models for moiré pattern detection in popular testing problems with lower computational costs. The most interesting thing is that just the direct usage of DRC coding results is able to provide satisfactory detection.</p>