【在 d******e 的大作中提到】 : 需要先作geometric transformation. : scaling, shifting,.... : a hard problem.
t*a
10 楼
谢谢大家回帖. 是啊,我前面帖子里已经作了 cross correlation,但是不能反映曲线的大小相差程度,即 index = Sum( (xi-average(X))*(yi-average(Y)) )/ sqrt( sum((xi-average(X))^2 ) * sum ((xi-average(Y))^2) ) 但是cross correlation不能反应两列值的大小相差程度,例如将X乘2,结果一样. 另外一个指标,"曲线拟合度"(coefficient of effciency) 能够较好反映出而列数的值的 大小关系 index = 1- sum((xi-yi)^2)/sum( (yi-average(Y))^2 ) 但是对于shape反映不好.我要做的曲线里除掉杂音后,一般有1到3个峰. 类似的曲线应该 峰的数量一致.
) 是Y,sorry
s*b
11 楼
you are dealing with a "clustering" problem. you can find this topic from many books on multivariable analysis or directly from some book on "clustrering". The definition of distance between two curves should incorporate with the problem you are working on. if you are sresearching on abstract curves, you may consider general distance, l1,l2...et al.. for a specific problem you may need specific wieght. Say, to emphasize "峰"的拟合 you may use weight abs(yi) or a increasing function respect to yi,