An image labeling mechanism using digital Radon projections

The digital Radon transform (DRT) maps any digital image into an equivalent set of independent digital projections. Digital projections with a low variance add nearly constant values to each pixel location in the back-projected image. The contents of low variance projections could be replaced with image descriptors or captions A simple mechanism is demonstrated to encode arbitrary messages into greyscale image data as digital projections. The inverse DRT distributes the projection data across the whole image, maximising the encoded message capacity and minimising the perturbation of original image pixel values. The method, similar in principle to spread-spectrum encoding techniques, does not require the original image to extract embedded messages, but is fragile. The DRT, being an additive operation, has a low calculation complexity and can be computed reasonably rapidly. Pixel sampling and interpolation issues, inherent for the processing of projections in continuous space, are avoided here as the DRT uses whole pixel contents to forward- and back-project digital projections. Rounding the data limits the encoded message capacity and recovery