An Accumulative Framework for Object Recognition

Abstract

Object recognition has received a lot of attention over the years and has reached a level where we have a lot of algorithms which can identify a large number of previously seen objects. We have algorithms which deal only with recognizing shapes and algorithms which are suited for recognizing objects in cluttered scenes using shape, color and texture. This dissertation provides a unified framework which can be applied not only to recognize simple shapes such as silhouettes but also recognize real objects in cluttered environments with occlusion.

The framework presented in this dissertation uses an accumulative approach reminiscent of the well known Generalized Hough Transform introduced by Ballard to recognize general shapes. Accumulator-based methods are highly parallel and use simple arithmetic. Noise and isotropic distortions tend to average out. The algorithm is invariant to translation, rotation, scale (zoom) and robust to illumination changes, background clutter, occlusion as well as view point changes. This is demonstrated using a wide range of data sets and experiments where it is shown to significantly outperform the current state-of-the-art.

The novel contributions of this dissertation are as follows: 1. A unified matching algorithm which matches different object models by accumulating features in a higher dimensional space. 2. A general and unified object representation (or model) built using robust and invariant features, extracted based on the nature of the object.