The Research Of Content-based Image Retrieval Techniques

With the development of internet and multimedia technologies, the rapid growth in the number of digital images, a large number of images circulated on the Internet, how to fast-growing mass of image data for fast, efficient retrieval is a hot issue in the current image applications. Therefore, the content-based image retrieval technology came into being, and become the research focus of the image retrieval technology.Content-based image retrieval are mostly concentrated in the global low-level features of the image, contains large amounts of information retrieval method based on the overall situation for some background, the user’s region of interest in the image smaller proportion of image retrieval results are less than ideal. To overcome the background of the image information for image retrieval based on regions of interest features outstanding image retrieval, but the question is how accurate the extracted region of interest.This paper interested in the color features and shape features a combination of image retrieval method, the first image corresponding split the region of interest, and then extract the color features and shape features of the target of interest and retrieve.Image are interested in the object extraction stage, the use of Fuzzy C-Means clustering algorithm to preprocess the image to reduce image noise and texture detail, so as to achieve the purpose of smoothing the image, and smooth the image at the same time ensure the image edge not be weakened, the segmentation accuracy; then watershed algorithm to segment the pretreatment images, and segmentation of images by the merger of the neighboring region of the merger guidelines. The experiments show that the algorithm can extract the image region of interest. Complete and clear the edge of the region of interest to improve the performance of image retrieval.Feature extraction in the region of interest is divided into color feature extraction and shape feature extraction. In the color feature extraction stage, the main colors of the K-means method to extract regions of interest to mention to remove the main colors of the color features; in the shape feature extraction stage, used to the heart of the chain code method to extract the characteristics of the direction of the boundary of the regions of interest, but the two objects of different shapes may also have the same chain code histogram, in order to solve this problem, to the heart of the chain code method has been improved, and spatial characteristics in the cardiac chain code string. After feature extraction, respectively, normalized color features and shape features external to the ultimate combination of color features and shape features of the similarity distance formula.The experiments show that the image retrieval method in the feature extraction to reduce the interference of the background area, to improve the retrieval accuracy.