Research, Image Retrieval Algorithm Based On Color And Texture Features

More and more digital images and video are being captured and stored. In order to use this information, content based image retrieval techniques were proposed. The main features used for image retrieval are color, texture and shape. This thesis looks into image retrieval techniques based on color and texture.First, we discuss content-based image retrieval techniques, including feature extraction, similarity measure, and evaluation criteria. Color space, color quantization and color feature extraction are the main techniques in image retrieval based on color feature. The existing color feature extraction methods include color histogram, accumulative color histogram, color moments. The experimental results using these three methods for image retrieval are showed in this thesis. Because these methods lack spatial information, someone divided the images into fixed geometrical regions and given different weights for each region to designate its importance. Image retrieval using fixed region segmentation is sensitive to rotations and absolute, relative spatial locations. As a result, image retrieval based on dominant color identification in the foreground and background of an image is proposed. This method reduces the dimensions of feature vector, incorporates spatial information, and is insensitive to rotations and absolute, relative spatial locations. And it is the central theme of this thesis. In this method, we look into the methods of image segmentation, the initial clustering center and their effects to the image retrieval's performance. In addition, we compare the retrieval performance using image retrieval techniques such as dominant color, equal-area segmentation, equal-area ring segmentation, dominant color identification in the foreground of an image, dominant color identification in the foreground and background of an image. The experimental results show that the techniques based on dominant color identification in the foreground and background of an image gets higher Precision and Recall rates than other existing image retrieval techniques based on color.In addition, we discuss the ways to describe texture features. In this thesis, gray-level co-occurrence matrix is used for image retrieval. And the experimental results using gray-level co-occurrence matrix for image retrieval are given in this thesis. Because we can hardly get better performance using one feature of an image, we combined the color and texture features for image retrieval. From the experimental results,we get higher Precision and Recall rates compared to the existing image retrieval techniques. In this thesis, all the methods for image retrieval are performed. From the experimental results, the method based on dominant color identification in the foreground and background of an image is very useful compared to the other methods in this thesis.