Home Projects Scholarships Curriculum SFS People


Featured Project:

PCAP Compression

Research Opportunities
Contact Us

percephashPerceptual Hashing

Authenticity is always an important issue as long as the multimedia contents are also available to other parties besides the owner. In the context of cryptography, the authenticity of the data is provided by cryptographic hash functions (MD5, SHA-1) in which the data is mapped to short bit strings that make up the hash value, which is then attached to the data. Whenever one would like to test the authenticity of the data, one simply recalculates the hash value from the data itself and compares it to the attached hash value.

Perceptual hash functions are designated hash functions for multimedia contents. Similar to cryptographic hash functions, they are required to generate different hash values for different inputs. However, here the definition of difference is changed from bitwise difference to perceptual difference. In other words, cryptographic hash functions generate a totally different hash value even if the input is changed by a single bit, while robust hash functions are expected to change the hash value only if the input is perceptually changed. For instance, the hash value of an image and its JPEG compressed version should be the same since they have no perceptual difference although their bit-string representation is completely different.

A good perceptual hash function should be: 1) robust: Manipulations that don't change the perceptual information should not change the hash value. 2) unique: Perceptually different inputs should have completely different hash values. 3) secure: It should be very hard to find (forge) perceptually different inputs having similar hash values.

Apart from authentication, perceptual hash functions can be used in applications where the multimedia content must be identified efficiently, such as in database indexing, broadcast monitoring, etc.

In ISIS we are designing perceptual hash functions for multimedia contents with maximized robustness, uniqueness and security.


Nasir Memon
Baris Coskun


Back to Projects