Research On Image Search Method Based On Hypergraph

Image retrieval is a hot topic in the field of information processing,it is widely used in pattern recognition,machine vision,statistical learning and many other research fields.Image retrieval can be described as follows: for a given image database and a query image,visual features and semantic features of these images are extracted by feature extraction algorithms.Then image retrieval system calculates the similarities between query image and images in the database and provides images with the high similarity to users.This paper does some research about visual features and semantic features of the image,explores some major problems in image retrieval and proposes multiple image retrieval algorithms.The specific content of this paper is shown as follows.This paper proposes a color image retrieval algorithm based on the hypergraph and integration framework.On the basis of this algorithm,this paper proposes a color image retrieval algorithm based on the hypergraph combined with a weighted adjacent structure.Using the two algorithms mentioned above,this paper uses semantic information to construct a hypergraph and proposes a color image retrieval algorithm based on the semi-supervised hypergraph.At first,this paper proposes a color image retrieval algorithm based on the hypergraph and integration framework.This algorithm proposes an integration framework to combine color difference histogram and micro-structure descriptor to calculate the similarity between images,which can better express the visual information of the image.Then,this algorithm uses the similarity between images to form a similarity matrix and construct a hypergraph by using the similarity matrix.This algorithm utilizes structural information of the hypergraph to find similar images.Compared with other retrieval methods,this method shows better retrieval precision.Then,this paper proposes a color image retrieval algorithm based on the hypergraph combined with a weighted adjacent structure.Color difference histogram,micro-structure descriptor and integration framework are used to calculate similarity between images in the database and form a similarity matrix.Then,this algorithm proposes a weighted adjacent structure to depict relationship between images by using the information of adjacent images.The weighted adjacent structure recalculates the similarity between images and updates the similarity matrix.By using the updated similarity matrix,this algorithm constructs a hypergraph and apply it to image retrieval.The proposed method is compared to other methods in several datasets.Experimental results manifest the performance and robustness of this proposed method.Finally,this paper proposes a color image retrieval algorithm based on the semi-supervised hypergraph.This algorithm uses the relevance feedback technology to get the label of the image pair(The two kinds of labels are divided into similar type and dissimilar type.)from users' operations and save the labels to a database.After that,color difference histogram,micro-structure descriptor and weighted adjacent structure are used to construct an initial hypergraph.Then this algorithm extracts some similar image pairs and dissimilar image pairs labels from the database as semantic information and integrate them into the construction of the hyperedge to form a semi-supervised hypergraph.The semi-supervised hypergraph is used to retrieve similar images.Experimental results manifest that the semi-supervised hypergraph can effectively improve the performance of image retrieval.