Skip to main content
. 2022 Apr 18;76:103703. doi: 10.1016/j.bspc.2022.103703

Table 2.

Haralick’s features description [35].

Haralick’s feature Equation Description
Angular second moment F1=i=1Ngj=1NgP(i,j)2 measures homogeneity of local gray scale distribution
Contrast F2=i=1Ngj=1Ngn2P(i,j) |i-j|=n, describes quantity of local changes happening in an audio
Correlation F3=i=1Ngj=1Ng(ij)P(i,j)-μxμyσxσy μx, μy, σx, and σy are mean and standard deviations, value lies between -1 to +1
Variance F4=i=1Ngj=1Ng(i-μ)2P(i,j) μ=i=1Ngj=1NgiP(i,j), measures spread of the signal
Inverse difference moment F5=i=1Ngj=1NgP(i,j)1+(i-j)2 measures local homogeneity
Sum average F6=r=02Ng-2rPx+y(r) measures mean of the gray level sum distribution
Sum variance F7=r=02Ng-2(r-F6)2Px+y(r) calculates dispersion of the gray level sum distribution
Sum entropy F8=-i=02Ng-2Px+y(r)log(Px+y(r)) measures disorder related to the gray level sum distribution
Entropy F9=-i=1Ngj=1NgP(i,j)log(P(i,j)) measure randomness
Difference variance F10=r=0Ng-1(r-l=0Ng-1lP|x-y|(l))2P|x-y|(r) describes heterogeneity
Difference entropy F11=-r=0Ng-1P|x-y|(r)log(P|x-y|(r)) measures disorder related to the distribution of gray scale difference
Information measure correlation 1 F12=F9-Hxy1max{Hx,Hy} Hx and Hy are the entropies of px and py respectively, and Hxy1=-i=1Ngj=1NgP(i,j)log(px(i)py(j))
Information measure of correlation 2 F13=1-exp(-2(Hxy2-F9)) Hxy2=-i=1Ngj=1Ngpx (i) py(j)log(px(i)py(j))
Maximum correlation coefficient F14=2ndlargesteigenvalueofQ(i,j) Q(i,j)=rP(i,r)P(j,r)px(i)py(k)