Hi. What are the alternatives we have to cross-correllation ? If I want to find degree to similarity between two sampled signals what options do I have besides doing the obvious correlation?? Thanks in advance
alternatives to correlation
Started by ●March 22, 2010
Reply by ●March 22, 20102010-03-22
On Mar 22, 2:44�pm, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> wrote:> Hi. What are the alternatives we have to cross-correllation ? > If I want to find degree to similarity between two sampled signals what > options do I have besides doing the obvious correlation?? > > Thanks in advanceTake a look at the amplitude magnitude difference function
Reply by ●March 22, 20102010-03-22
sanam1 wrote:> Hi. What are the alternatives we have to cross-correllation ? > If I want to find degree to similarity between two sampled signals what > options do I have besides doing the obvious correlation??Define "similarity". Depending on this, there could be infinitely many ways to measure it. VLV
Reply by ●March 22, 20102010-03-22
On Mar 22, 3:54�pm, maury <maury...@core.com> wrote:> On Mar 22, 2:44�pm, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> > wrote: > > > Hi. What are the alternatives we have to cross-correllation ? > > If I want to find degree to similarity between two sampled signals what > > options do I have besides doing the obvious correlation?? > > > Thanks in advance > > Take a look at the amplitude magnitude difference functionAny reasonable alternative measure should be between signals that are normalized for differences in scale and location. Hope this helps. Greg Gre
Reply by ●March 22, 20102010-03-22
Greg Heath wrote:> On Mar 22, 3:54 pm, maury <maury...@core.com> wrote: > >>On Mar 22, 2:44 pm, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> >>wrote: >> >> >>>Hi. What are the alternatives we have to cross-correllation ? >>>If I want to find degree to similarity between two sampled signals what >>>options do I have besides doing the obvious correlation?? >> >>>Thanks in advance >> >>Take a look at the amplitude magnitude difference function > > > Any reasonable alternative measure should be between signals that are > normalized for differences in scale and location.Extend this to any affine transformations ? VLV
Reply by ●March 23, 20102010-03-23
Hi, Thanks for your reply. Suppose if I have two signals and I correlate them then I get degree of similarity between them. But my question is that I want to achieve same result but with an operation other than correlation. Hope I have made my question clear. Regards, Sanam>sanam1 wrote: > >> Hi. What are the alternatives we have to cross-correllation ? >> If I want to find degree to similarity between two sampled signals what >> options do I have besides doing the obvious correlation?? > >Define "similarity". >Depending on this, there could be infinitely many ways to measure it. > >VLV >
Reply by ●March 23, 20102010-03-23
Hi, How can I use affine transformation here??> > >Greg Heath wrote: > >> On Mar 22, 3:54 pm, maury <maury...@core.com> wrote: >> >>>On Mar 22, 2:44 pm, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> >>>wrote: >>> >>> >>>>Hi. What are the alternatives we have to cross-correllation ? >>>>If I want to find degree to similarity between two sampled signalswhat>>>>options do I have besides doing the obvious correlation?? >>> >>>>Thanks in advance >>> >>>Take a look at the amplitude magnitude difference function >> >> >> Any reasonable alternative measure should be between signals that are >> normalized for differences in scale and location. > >Extend this to any affine transformations ? > >VLV >
Reply by ●March 23, 20102010-03-23
On Mar 23, 8:44�am, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> wrote:> Hi. What are the alternatives we have to cross-correllation ? > If I want to find degree to similarity between two sampled signals what > options do I have besides doing the obvious correlation?? > > Thanks in advanceYou could shift one wrt the other, multiply and sum.
Reply by ●March 23, 20102010-03-23
On Mar 23, 2:10�am, HardySpicer <gyansor...@gmail.com> wrote:>> You could shift one wrt the other, multiply and sum.And what is the difference between this shift-multiply-sum operation, by which you presumably mean something like sum_{over all i} x[i]*y[i+k] and the cross-correlation that the OP does not want?
Reply by ●March 23, 20102010-03-23
On Mar 23, 2:29�am, "sanam1" <sanamsingh@n_o_s_p_a_m.hotmail.com> wrote:> Hi, > Thanks for your reply. Suppose if I have two signals and I correlate them > then I get degree of similarity between them. But my question is that I > want to achieve same result but with an operation other than correlation. > Hope I have made my question clear. > Regards, > Sanam > > >sanam1 wrote: > > >> Hi. What are the alternatives we have to cross-correllation ? > >> If I want to find degree to similarity between two sampled signals what > >> options do I have besides doing the obvious correlation?? >What are you trying to accomplish and why is correlation not sufficient ?? You want something that gives the same results as the correlation but by some other operation, so I am just trying to understand why. You could also use any metric distance if you are dealing with a hilbert space at treat your signals as vectors. The dot product (correlation at 0 shift) is a classic example. Distance metrics are monotonic functions of similarities you can use Manhantan, Mahalanobis (if dealing with distributions), or other any of the p-vector sum distances .... There is also coherence analysis were similarity is computed in spectral domain (used frequently in eeg analysis).
