An improved scale invariant feature transform algorithm based on rounded projection

Abstract

The scale-invariant feature transform algorithm proposed by Lowe has a low efficiency which can not meet the need of real-time. The algorithm based on rounded projection proposed in our paper applies Fast Fourier Transform algorithm (FFT) on the projection local area to compute the first harmonic components which are used to prescreen the feature points that extracted by SIFT algorithm. We carry on the image to match according to the prescreened feature points to calculate the local area descriptors. The experiments show that new algorithm has a less number of feature points than the original, so it improves the efficiency and has a better performance.