Automatic synthesis of image details based on multiresolution coherence

This paper introduces a new kind of image coherence based on a correlation of the spatial distribution of image gray values across successive resolutions or scales. By capturing this property statistically from an image class, a reference base is built for automatic synthesis of fine image details in any image from the class. We exploit such multiresolution spatial coherence to synthesize fine details of an image at a high resolution from a low-resolution input image. We demonstrate that this approach approach is particularly suited to synthesizing details of calligraphy.