Abstract: Content-based remote sensing image retrieval has become a research hotspot in remote sensing field. In view of this, a new method based on this bag of visual words model and color histogram is proposed for remote sensing image retrieval. The method extracts image local invariant features with scale invariant feature descriptor, combines local features by bag of visual words model, and generates pyramid histogram for each image. Then a more distinctive feature vector is achieved by combining the color histogram of each image, the support vector machine classifier is trained using the feature vector set generated last step, and the images classified into one category with the query image then to be output through the classifier. Finally remote sensing image retrieval procedures are completed. The experimental results show that the proposed method not only improves the precision and recall of image retrieval, but also verifies that the method can efficiently overcome the changes of illumination, noise and direction, and has better robustness.

Key words: local invariant features, bag of visual words model, color histogram, support vector machine classifier, image retrieval