Empirical analysis of SIFT, Gabor and fused feature classification using SVM for multispectral satellite image retrieval

High Level image understanding and Content extraction is becoming a challenging task in Content based image retrieval system for satellite images. Retrieval based on the low level extraction techniques does not bridge the semantic gap. In the experiment high level feature extraction techniques i.e. scale invariant feature transform and Gabor descriptors are used. The novel approach is proposed in which both the feature descriptors are fused to retrieve the results with more accuracy rate. The experiment is conducted on the multispectral satellite images, of Landsat 8 sensor. The similarity of the query image to that of stored database images is matched by the Manhattan distance. The Precision and Recall is computed for the data set. The results have shown the improved retrieval rate. The retrieval efficiency is further increased by using the SVM classifier by classifying the satellite images based on Urban area, Water body and Vegetation. The experimental results shows that the fusion technique gives better result and more accuracy can be obtained by classifying the dataset using SVM.