Catalog | Conferences | - Departments - | First Year | Handbook | Honors | Learning Spaces | Off Campus | Research | Support | Writing Center

Mathematics and Computer Science

Kernel Methods for Image Processing

Dan Bucatanschi

Honors Thesis, 2006

Mathematics and Computer Science Department

Denison University

Advisor: Matt Kretchmar

Abstract: We discuss the possibility for a computer to identify people and human features, such as long hair and smiles, from photographs of human faces. These problems fall into the field of supervised learning. We begin with a discussion of the broader problem of classification and various ways in which it is traditionally solved. We introduce the kernel and its use in solving problems. We then present various algorithms and the methods they employ together with kernels to solve the supervised learning tasks. We finish our discussion by presenting the results and conclusions of solving three tasks: identifying a person out of photographs of several people, detecting photographs of people with long hair, and finding photographs of smiling people.