Research On Image Retrieval Algorithms Based On Global And Local Features

With the development of network and digital technology, the amount of images has grown rapidly. It is important to retrieve suitable images from the internet or particular image collections efficiently, and image retrieval has become a problem worthy of study. Theoretically, three methods are mentioned in image retrieval area: text-based, content-based and semantic-based. Text-based image retrieval(TBIR) is very sensitive to the key words labeled by people and it has relatively lower efficiency and accuracy, due to the artificial marked subjectivity. Content-based image retrieval(CBIR) systems overcome these difficulties. It has become a new hotspot. CBIR systems extract the low-level image features and process them in a specific way. These features include color, texture, shape and space structure feature. These features can present the content of an image. The basic process of this method is that users provide an example image firstly. The systems extract the low-level features and compute the similarity between the query image and images in databases. Finally the top images are shown to users. The rapid development of the CBIR method has basically solved different problems in image retrieval area. The main research direction of this paper is concerned on image retrieval method based on the integration of global and local features. The main work of this paper is as follows: 1. Image Retrieval Method Based On The Feature Associated FusionThis paper proposes a new image retrieval method based on color and texture features of the interrelated fusion for content-based image retrieval(CBIR). First, HSV color space is quantified rationally to extract the color features. In the same time, we extract image texture features by using wavelet transform. Because of the complexity of images, we adopt two methods in this paper. When the contents of images are simple, we adopt the serial correlation method; For the complex retrieval images, we adopt the parallel correlation method based on canonical correlation analysis(CCA). The experimental results show that this method is feasible. 2. Image retrieval method based on the integration of global and local featureImage retrieval can be divided into the following steps: image segmentation, feature extraction, calculation of the global features and local features and feature similarity measure. Among them, the extraction and computation of the global and local features are the key steps in the process of image retrieval. In order to achieve a variety of features in the process of image retrieval, global features adopt the global color histograms and wavelet texture features and local features use the method of traversing local color histogram. We measure the distance between the features of images, then the image retrieval results are shown to users.