SynMoire

Addressing the limitations and inconveniences imposed by the randomness in moiré pattern generation on deep learning model training, we have constructed the SynMoiré dataset through a synthetic approach to generate moiré images. The construction process involves resampling the original images into an RGB sub-pixel format, applying random projection transformations, radial distortions, and Gaussian filtering to simulate camera effects. Subsequently, the images are resampled using a Bayer color filter array, followed by demosaicing and JPEG compression with a quantization factor of 0.5 to induce moiré patterns. To optimize color and brightness, adjustments are made to the Cb and Cr channels in the YCbCr color space based on the original images. Compared to other public datasets, SynMoiré exhibits moiré effects that are closer to real-world scenarios, providing robust support for model training.