Clustering and Symbolic Signature Classification for Image Retrieval

by
Jenny Yuen

When we look around, we categorize objects in an unconscious level. We can tell there is a house, a building, or the sky. But the conceptualization of an object as a house, a building, or the sky depends on various factors. There can be many different types of houses, depending on the geographical zone. A skyscraper is a building but an old medieval castle is, too. The sky is different on a rainy day compared to a sunny one, or at night. If we are talking about images with objects, an object can be different from others in size and view point, the object can be partially occluded by another, or the image might only capture a part of the object. Therefore, these barriers of ambiguity and diversity are obstacles to finding an easy solution for automated recognitions of objects in images. The purpose of this research is to automate object recognition of a finite set of known categories by analyzing features like color, texture, and structure in an image and introducing symbolic signatures in the representation of objects found in an image.

Advised by Linda Shapiro

MGH 251
Wednesday
February 4, 2004
3:30 - 4:20 pm