Fast High-dimensional Feature Matching for Retrieving Remote Sensing Images

The key point in applying high-dimensional local features to remote sensing image retrieval is to improve the efficiency of feature matching. A new Compressed Priority Filter (CPF) algorithm is investigated that quantizes the feature vectors to compress the search space, constructs a high-dimensional index, searches candidates via priority queue, and calculates the exact feature vectors to get nearest neighbors. Then, a fast remote sensing image retrieval algorithm based on Speeded Up Robust Feature (SURF) features is proposed based on CPF. It is proved by experiments and via analysis that CPF can reduce disk I/O and float-pointing calculation. When the number of features is big, it is much faster and more precise than the classical BBF algorithm. It is obvious that the fast remote sensing image retrieval algorithm based on SURF can return to the correct related target image from the gallery quickly, together with similar images.