Texture search engine

Abstract

Texture, an important element in computer graphics, is being widely used today to facilitate the visualization of complex concepts and ideas. With the enormous amount of texture images available, there is a need for an efficient way to search for and retrieve relevant texture images.

The focus of this project is to improve the speed of a prototype texture search engine. The prototype uses sketches instead of text as inputs for queries, inspired by the idea of Content-based Information Retrieval which focuses on the content of images over descriptive text. The main technique used in this report combines shape context with feature tracking in searching for similar patterns. To improve the efficiency of the prototype, a factoring approach was also introduced. The factoring approach managed to improve the search speed while maintaining a relatively high accuracy of 84% for the alphabetical test cases. In addition, a separate template matching method will also be discussed to provide a means for comparison.