两个直方图之间有很多距离量度。您可以在以下位置阅读这些措施的良好分类:
K. Meshgi和S.Ishii,Proc.National.Proc.Natl.Acad.Sci.USA中,“用网格扩展颜色的直方图以提高跟踪精度”。2015年5月,日本东京,MVA'15会议开幕。
为了方便起见,此处列出了最受欢迎的距离功能:
DL0=∑ih1(i)≠h2(i)
DL1=∑i|h1(i)−h2(i)|
DL2=∑i(h1(i)−h2(i))2−−−−−−−−−−−−−−−√
DL∞=maxi|h1(i)−h2(i)|
- L p或分数距离(Minkowski距离族的一部分)p
DLp=(∑i|h1(i)−h2(i)|p)1/p和0<p<1
D∩=1−∑i(min(h1(i),h2(i))min(|h1(i)|,|h2(i)|)
DCO=1−∑ih1(i)h2(i)
DCB=∑i|h1(i)−h2(i)|min(|h1(i)|,|h2(i)|)
DCR=∑i(h1(i)−1n)(h2(i)−1n)∑i(h1(i)−1n)2∑i(h2(i)−1n)2√
- 柯尔莫哥洛夫-史密尔诺夫(Kolmogorov-Smirnov Divergance)
DKS=maxi|h1(i)−h2(i)|
DMA=∑i|h1(i)−h2(i)|
DCM=∑i(h1(i)−h2(i))2
Dχ2=∑i(h1(i)−h2(i))2h1(i)+h2(i)
DBH=1−∑ih1(i)h2(i)−−−−−−−−√−−−−−−−−−−−−−−−−√和hellinger
DSC=∑i(h1(i)−−−−√−h2(i)−−−−√)2
DKL=∑ih1(i)logh1(i)m(i)
DJD=∑i(h1(i)logh1(i)m(i)+h2(i)logh2(i)m(i))
- 地球移动者的距离(这是将距离分类信息A嵌入距离中的运输距离的第一个成员,有关更多信息,请参阅上述论文或Wikipedia条目。
DEM=minfij∑i,jfijAijsumi,jfij
∑jfij≤h1(i),∑jfij≤h2(j),∑i,jfij=min(∑ih1(i)∑jh2(j))fijij
DQU=∑i,jAij(h1(i)−h2(j))2−−−−−−−−−−−−−−−−−−−√
DQC=∑i,jAij(h1(i)−h2(i)(∑cAci(h1(c)+h2(c)))m)(h1(j)−h2(j)(∑cAcj(h1(c)+h2(c)))m)−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−√00≡0
我的GitHub存储库中提供了其中一些距离的Matlab实现:https :
//github.com/meshgi/Histogram_of_Color_Advancements/tree/master/distance
您也可以搜索Yossi Rubner,Ofir Pele,Marco Cuturi和Haibin Ling之类的人更先进的距离。
更新:有关距离的替代解释出现在文献中的各处,因此为了完整起见,在此列出它们。
DCB=∑i|h1(i)−h2(i)||h1(i)|+|h2(i)|
DBC=1−2∑ih1(i)=h2(i)∑ih1(i)+∑ih2(i)
DIOU=1−∑imin(h1(i),h2(i))∑imax(h1(i),h2(i))