The Research Of Multi-feature Image Retrieval Algorithm Based On K-means Clustering Segmentation

With the widely used of digital equipment and rapidly development of internet technology, the image database has grow as geometric progression trend. How to find the required image rapidly and accurate in image database has become an urgent problem. In this context, the Content-based image retrieval technology came into being, and became a big research hot spot. Most mature retrieval technology are based on the image visual features, using color, shape and texture as feature vector for similarity measure and image retrieval. The image retrieval methods based on single feature often has low precision. Image retrieval methods based on multi-feature can improve retrieval precision. This paper focused on how to use the multi-feature for image retrieval. The main research contents and contributions are as follows:1. Check and study the relevant literature material of content-based image retrieval, summary the key technology of content-based image retrieval. First, this paper introduced the extraction and description method for color, shape and texture feature, and special for multi-feature. Then, it introduced the technology of similarity measure. At last, it introduced the performance evaluation method for image retrieval.2. A new approach for image retrieval using multi-feature based on K-means clustering algorithm is proposed. The algorithm firstly converted image to HSV space, The H component and V component were restructured and clustered by K-means clustering algorithm. Then take intersection operation for the two images. Target object was obtained. Concerning the shape information, Hu invariant moments and fourier descriptors of the target object were extracted. Then euclidean distance was adopted for similarity measure. The experimental results showed that the algorithm has a good segmentation effect and the retrieval results.3. Designed a content-based image retrieval system based on single computer. Firstly analysis the demand and design the system architecture. The system mainly has two modules, the feature database subsystem and image retrieval subsystem. The data was saved by file. Some sorts of different retrieval algorithms were tested and compared with this system. The Experiment results show that the new algorithm was effective and the system was feasibility.