A speeded-up affine invariant detector

Abstract

Local features extraction is one of the most important image processing algorithms, which can be widely used in object recognition, image registration, tracking, etc. To extract key points of an image, SIFT (Scale Invariant Feature Transform) is scale invariant and SURF (Speeded Up Robust Features) is more than three times faster than SIFT. But neither of them works well under large viewpoint angle change. ASIFT (Affine SIFT) is fully affine invariant thus working well when viewpoint angle is huge; but it cannot reach computational efficiency due to its complexity. A speeded-up affine invariant detector for local features extraction is proposed in this paper, which makes use of the affine invariance of ASIFT and the efficiency of SURF. Experimental results on image features extraction and matching demonstrate the robustness and efficiency of the proposed algorithm. Compared with the state-of-art methods, the proposed detector can keep affine invariant as ASIFT by only a quarter of its time consuming.