Feature Fusion Based Image Retrieval

The rapid development of image retrieval technology benefits from stable image features and Bag-of-features image representation model. However, during the image feature extraction and Bag-of-features representation process, there has a large amount of information loss, affect the accuracy of search results. To solve these problems, this paper researches the feature fusion to retain more of the original image information to improve the accuracy of image matching. The main contents of this paper are as follows:First, this paper reviews the research history of image retrieval technology and introduces the Bag-of-features based image retrieval model. By analyzing the shortcomings of the Bag-of-features model, this paper argues that the original image information missing is a key factor restricting the accuracy of image retrieval.Secondly, the chapter III presents color and SIFT fusion based image retrieval. In order to preserve more image information, the method couples the color information into the Bag-of-features model as a complement to sift features, two different features are extracted to represent the same interest region and two visual words are obtained by quantizing the descriptor pair with two independent codebooks. In addition, this method constructs a two-dimensional inverted index table to achieve feature fusion and efficient image retrieval. Experiments results on Holidays dataset demonstrate that the proposed approach significantly improve the performance of Bag-of-features model.Finally, the chapter IV presents a salient object based image similarity weighting strategy. The visual attention mechanism is adopted into Bag-of-features model to extract image salient objects, which express the semantic image information. In the image retrieval process, the salient objects are used to weighting the image similarity, so the image with the global similarity and also similar partial obtain a higher score. Experimental results show that the proposed similarity weighted strategy further improve the accuracy of image retrieval.