| 摘要: |
| 针对单一特征不能很好地表述图像的问题，提出了一种融合多特征的图像检索算法。首先，提取查询图像和图像库中样本图像的GIST（Generalized Search Tree)特征，用欧氏距离衡量图像间的GIST相似度值，根据查询图像的GIST特征在图像库中进行检索，将结果按相似度进行排序；然后，提取查询图像和返回结果中前k幅图像的尺度不变特征变换(SIFT)特征，使用BBF(Best Bin First)算法进行特征匹配；最后，通过特征点匹配点对数排序并返回检索结果。实验在改进的Corel1000数据集上进行，与传统的单特征图像检索算法对比，提出的图像检索算法不仅提高了检索准确率，而且获得了较好的检索效率。 |
| 关键词: 图像检索 图像匹配；GIST特征；SIFT特征；特征提取 |
| DOI：10.3969/j.issn.1001-893x.2017.09.008 |
|
| 基金项目:国家自然科学基金资助项目(61772561，61602528，61602529） |
|
| An image retrieval algorithm based on multi-feature fusion |
| QIN Jiaohua,XIE Bei,XIANG Xuyu,YU Wentao,YI Jizheng |
| (College of Computer Science and Information Technology,Central South University of Forestry and Technology,Changsha 410004,China ) |
| Abstract: |
| In order to solve the problem that the single feature cannot represent an image completely,an image retrieval algorithm based on multi-feature fusion is proposed. First,the generalized search tree(GIST) features of the query image and all the images in the image library are extracted. The similarity of two images is measured by the Euclidean distance. The image library is retrieved by the GIST feature of the query image,and the results are arranged by sorting the similarity value. Second,the scale invariant feature transform(SIFT) features of the query image are extracted as well as the k sub-images which are at the front of the returned results. Then,the feature matching is performed by best bin first(BBF) algorithm. Finally,the retrieval results are returned by sorting the number of matching points. The experiment is carried out on the improved Corel1000 dataset. Compared with traditional single feature image retrieval algorithms,the proposed algorithm not only improves the retrieval accuracy but also achieves better retrieval efficiency. |
| Key words: image retrieval image matching GIST feature SIFT feature feature extraction |
