The coefficient of coefficient of correlativity between two signals (cross correlation) is a streamer antenna to feature detection. The normalized form of cross correlation (normalized correlation coef?cient) is particularly used for template matching. In this case, the two-dimensional correlation of human bodys is considered. wholeness of its biggest drawbacks is the need for a lot of computer scienceal power, especially when or so correlation coef?cients are computed. This paper presents a new rule for a high performance thread- and data-parallel computation of normalized cross correlation in the spatial domain. It will be shown that a acceleration of up to 5 laughingstock be achieved solely by a sophisticated programming of the SIMD unit of a standard microprocessor. Furthermore, the new dataparallel implementation in the spatial domain can raze outperform an (also data-parallel) oftenness domain implementation.
1. Introduction
The correlation coef?cient is a robust measure for the similarity of two signals and is therefore a good deal used for template matching. Due to the high computational appeal of spatial domain convolution it is often computed in the frequency domain using the fast fourier transform.
While cross correlation can easily be expressed in the frequency domain, the normalized form of correlation does not have a correspondingly simple and ef?cient frequency domain expression. Although methods for an ef?cient frequency domain computation already exist [1], the spatial domain computation offers some advantages. Computation in the frequency domain has to be performed with at least 32 indorsement ?oating point values even if image and template consist of grey scale pixels with 8 bit each. In the spatial domain the 8 bit values can be used instead and the correlation coef?cient can be computed using less memory because image and template mostly ?t in the CPUs L1 or L2 cache. Thereby the number
of accesses to the main memory are signi?cantly lessen (compared to a computation in the frequency domain). The...If you want to personate a full essay, order it on our website: Orderessay
If you want to get a full essay, wisit our page: write my essay .
No comments:
Post a Comment