We present new algorithms for hashing, or one-way image compression, and comparison of bitmapped images. Our methods are based on random multiscale subdivision of images into regions, randomized rounding of intensity averages in those regions, and robust compression of the resulting vectors by error correction. As hashes useful for robust identification and comparison of images, the compressed vectors can replace watermarks. Our schemes work with images subjected to common distortions, including scanning, resizing, and resampling. Additionally, image hashes withstand anti-watermark transformations performed by software such as StirMark and unZign.