Normal Synthesis on RGBN Images

In this work, we synthesize normals and color to add geometric details to an RGBN image (image with a color and a normal channel). Existing modeling and image processing tools are not apt to edit RGBN images directly. Since high resolution RGBN images can be obtained using photometric stereo, we used them as full models and as exemplars in a Texture from Example synthesis. Our method works on RGBN images by combining the normals from two bands: base shape and details. We use a high pass filter to extract a texture exemplar, which is synthesized over the model’s smooth normals, taking into account foreshortening corrections. We also discuss conditions on the exemplars and models that guarantee that the resulting normal image is a realizable surface.