Automatic Enhancement of Handheld Photographs

This internship proposes to create a support vector machine based classifier which classifies image aesthetics, ranking images from 1-7. An existing classifier has been developed which classifies professional quality photographs. To be of use to Vidigami, the classification method needs to be developed which can provide an aesthetics score for average quality photographs. Individuals taking photos within the context of Vidigami’s service generally are not professional or semi-pro photographers, and are unlikely to produce a significant number of high quality photos in the general aesthetic sense. In addition to the aesthetic ranking of images from 1-7, a number of other general image properties are desirable to detect. If we can detect blurry photos, over/underexposed photos, or photos that are nearly identical, these can be pre-filtered out of the selection process. Finally, in order to be of use to Vidigami, the development of a prototype photo selection system which select N photos from a set of M is critical. Creation of such a system is complex, however exploring the use of the classifiers described above as elements begins to give us an idea of how to go about its development.

Faculty Supervisor:

Dr. Sid Fels

Student:

Steve Oldridge

Partner:

Vidigami Media Inc.

Discipline:

Engineering

Sector:

Information and communications technologies

University:

University of British Columbia