Parallel scale space construction using SIMD hypercube

Abstract

This paper proposes parallel scale space construction of Scale Invariant Feature Transform (SIFT) using SIMD hypercube. The parallel SIFT approach is used for iris feature extraction. The input iris images and Gaussian filters are mapped to each processor in the hypercube and convolution takes place in each processor concurrently. The time complexity of parallel algorithm is O(N 2) whereas sequential algorithm performs with complexity of O(lsN2), where l is the number of octaves, s is the number of Gaussian scale levels within an octave for N2 sized iris image. � 2012 IEEE.