Texture Segmentation by Genetic Programming

We investigated image retrieval using texture segmentation by genetic programming. In this study, we are interested with two textures: sky and grass textures. Single-step texture classification by genetic programming was used. Based on the result of texture segmentation, an image will be labelled as having the textures of interest or not. Then, the image retrieval system returns the images which are labelled as having the textures of interest. The image retrieval accuracies were quite good for sky texture: 95% and 84% for training and testing databases. For grass texture, it was able to achieve 75% and 83% for training database and testing database respectively. The images in the databases are taken from Corel image set volume twelve which the images are general purpose images. It was observed that texture can be used for image retrieval for relatively simple texture such as sky texture. Further investigations would be needed for grass texture and more textures. 1 

• Type: Journal
• Year: 2008
• Where: EC
• Authors: Andy Song, Victor Ciesielski
• Keywords: EC 2008, ECommerce, Image, Image Retrieval, Texture Segmentation