Research And Application Of Clustering Based On Semantic Image Retrieval

With the development of the Internet and multimedia technology, digital image resources become rich, how to quickly and efficiently retrieve images that meet users’ needs from the mass of the image database is a very important research now. Users pursue more high-level semantic similarity and matching when retrieving the image, so the image retrieval based on semantic gradually becomes the focus of research. Feature extraction of the bottom of images has been relatively mature, while the extraction and description technologies of the high-level semantic of images are still difficult problems, therefore, to establish the connection between low-level features and high-level semantic has become the key of the "semantic gap" problem.This paper presents a semantic image retrieval method based on clustering, which establish the connection between the low-level visual features and high-level semantics. Clustering is an unsupervised learning method, and it has been used in the field of image segmentation, not directly used for image retrieval, but in this paper clustering is applied to image retrieval. First the characteristic matrix cluster image is established based on features, the connection between low-level features and high-level semantic is established to divide images into different semantic categories; then comparison of the similarity. Those make the retrieval results more consistent with users’needs, and make retrieval efficiency improved.This paper compares two different clustering methods under the application of image retrievals. Color features of the image are extracted, clustering in the image database is carried out by means of traditional K-means clustering method and improved NCut method, and the results were analyzed. The experiments have shown NCut clustering method can get better results, and when it is applied to image retrieval, better retrieval results can be obtained.This article uses the NCut clustering, image color feature and shape feature, designs a semantic retrieval system based on clustering. First, we extract the image color feature., and then establish the characteristic matrix clustering image, put the images into different semantic categories, calculate the similarity between query image and each sub-class, extracting the most similar sub-class image’s shape feature, calculate the similarity between query image and each image which in the sub-class, get search results, in which content-based image retrieval, shape-based image retrieval is better. Then the proposed clustering-based semantic image retrieval system is compared with the typical shape-based image retrieval system. By the experimental analysis it’s shown that my retrieval system has a good retrieval effectiveness, recall and precision increased.