pHash

Verification & Validation

Preliminary results based on a Matlab prototype showing the peaks of cross correlation (PCC) between the original image and its attacked version for the RADISH algorithm. The key point is that the PCCs are all above a threshold of 0.90. For the DCT/pixel tube algorithm, the hamming distances are all under a threshold of 22 for matching images, while non-matching images have a distance greater than 22.

Intra PCC results for 60% compression, blur, 80% and 120% scaling attacks, and 4-pixel off-center with 3 degree rotation attacks.

Inter PCC results for original image and completely different image.

Hamming distance plot of image and its attacked version, followed by hamming distance plot between original image and completely different image.

The next set of images show the intra and inter PCC values of the Histogram-based feature vector. While the intra PCC values are all above a threshold ~0.84, the inter PCC values show no clear threshold, indicating the histogram based approach would lead to a large number of false positives (e.g. images said to be a match but are truly not).

Intra PCC results for the histogram-based feature vector.

Inter PCC results for histogram-based feature vector.

Reference

B. Coskun, B. Sankur, Robust Video Hash Extraction, EUSIPCO.2004: European Conf. On Signal Processing, Vienna, September 2004.

C. De Roover, C. De Vleeschouwer, F. Lefebvre, B. Macq, Robust Image Hashing Based on Radial Variance of Pixels, ICIP, volume 3, pages 77-80, Genova, Italy, September 11th-14th, 2005.

François-Xavier Standaert, Frédéric Lefèbvre, Gaël Rouvroy, Benoit M. Macq, Jean-Jacques Quisquater, Jean-Didier Legat: Practical Evaluation of a Radial Soft Hash Algorithm. ITCC (2) 2005: 89-94.