Research And Realization Of Index In The Content-based Image Retrieval

Recently, the micro-electronic, communication and computer technique have been greatly developed. Because of the enhancement of memory technique and digital image equipment, and the popularity of Internet, people have much more image resources nowadays. But how to effectively and efficiently retrieve images has become an urgent problem.Because of the complexity of images, the traditional image retrieval techniques, which are based on text description and keyword, are not effective. So the content-based image retrieval (CBIR) has been widely reseached. CBIR directly extracts image feature, and retrieval image by comparing similarity. If the method of feature extraction and similarity measure is chosen, the rest work is to retrieve most similar images in image database.The simplest method is sequentially scanning, measuring the similarity between the query example and each image in database. But the computational effort of feature similarity is very large, sequential scanning is inefficient. Hence, the effective and efficient index is needed for CBIR.Because of the speciality of the image feature, the practicable index technique for feature data is spatial index. This paper researches the widely used spatial index such as R-tree and SS-tree. The idea of improve these index using cluster is introduced. A new index, called CICR-tree, is proposed to organize and search dynamic high-dimension vast image data set in CBIR. The CICR-tree is a multi-way balance tree, which is based on SS-tree and K-means cluster. A Web-based CBIR system and these indices is realized using Java.The experiment shows that with the use of index, the efficiency of CBIR is greatly improved. And the performance of CICR-tree is better than present index such as R-tree, SS-tree. Using index to retrive is efficient and effective for CBIR.