Some existing forensic tools allow a fast recognition of known illegal pictures on a hard disk. Most of the time, these tools are based on a cryptographic one-way hash function (typically the MD5) that generates a hash value for any image. The hash value is then used to identify the image; it plays the role of a "fingerprint" for the image.
With such a method based on a cryptographic one-way hash function, a small transformation applied to an image will change its hash value drastically. We will present recent results of ongoing research the objective of which is the development of a more robust recognition algorithm. This new algorithm should be able to cope with the deformations commonly applied to images, so that the fingerprint of an original image often corresponds to the one of the transformed image. We have focused our research on the use of information contained in the histogram of an image. The algorithm consists in a series of operations on the histogram itself in order to extract relevant information that allows a robust comparison of images.