GLRLM,全程 gray-level run-length matrix,这里翻译为 灰度形成矩阵。
概念描述: 灰度共生矩阵用于描述图像某一像素值 i i i沿某一方向 θ \theta θ连续相邻为 j j j的个数,记GLRLM中一个值为 D ( i , j , θ ) D(i,j,\theta) D(i,j,θ),其中 i i i是像素值或灰度值, j j j是连续相邻数, θ \theta θ是角度或方向,一般 θ \theta θ的值为 0 o , 4 5 o , 9 0 o , 13 5 o 0^o,45^o,90^o,135^o 0o,45o,90o,135o。
举例说明: 下图图(a)表示一副图像的灰度值,因为 4 × 4 4 \times 4 4×4矩阵,故在 0 o 0^o 0o方向上某元素最大连续相邻数只能为4。
所以,在 0 o 0^o 0o方向上的GLRLM为: [ [ 1 , 2 , 0 , 0 ] , [ 0 , 2 , 0 , 0 ] , [ 0 , 1 , 1 , 0 ] , [ 0 , 1 , 1 , 0 ] ] [ [1,2,0,0], [0,2,0,0], [0,1,1,0], [0,1,1,0] ] [[1,2,0,0],[0,2,0,0],[0,1,1,0],[0,1,1,0]]
0 o , 9 0 o , 13 5 o , 4 5 o 0^o,90^o,135^o,45^o 0o,90o,135o,45o方向上的GLRLM分别如图(b ),(c ),(d ), (e )所示。
像素值(灰度值)为 i i i,连续相邻数 j j j,角度值(方向)为 θ \theta θ, p ( i , j ∣ θ ) p(i,j|\theta) p(i,j∣θ)为在 θ \theta θ方向上连续 j j j 个 i i i值的情况的个数或概率(用频率近似概率)。
则有:
1.Short Run Emphasis(SRE)
S R E = ∑ i ∑ j [ p ( i , j ∣ θ ) j 2 ] ∑ i ∑ j p ( i , j ∣ θ ) SRE=\frac{\sum_i\sum_j[\frac{p(i,j|\theta)}{j^2}]}{\sum_i\sum_jp(i,j|\theta)} SRE=∑i∑jp(i,j∣θ)∑i∑j[j2p(i,j∣θ)]
2.Long Run Emphasis(LRE)
L R E = ∑ i ∑ j j 2 p ( i , j ∣ θ ) ∑ i ∑ j p ( i , j ∣ θ ) LRE=\frac{\sum_i\sum_jj^2p(i,j|\theta)}{\sum_i\sum_jp(i,j|\theta)} LRE=∑i∑jp(i,j∣θ)∑i∑jj2p(i,j∣θ)
3.Gray Level Non_Uniformity(GLM)
G L M = ∑ i [ ∑ j p ( i , j ∣ θ ) ] 2 ∑ i ∑ j p ( i , j ∣ θ ) GLM=\frac{\sum_i[\sum_jp(i,j|\theta)]^2}{\sum_i\sum_jp(i,j|\theta)} GLM=∑i∑jp(i,j∣θ)∑i[∑jp(i,j∣θ)]2
4.Run Length Non_Uniformity(RLM)
R L N = ∑ j [ ∑ i p ( i , j ∣ θ ) ] 2 ∑ i ∑ j p ( i , j ∣ θ ) RLN=\frac{\sum_j[\sum_ip(i,j|\theta) ]^2}{\sum_i\sum_jp(i,j|\theta)} RLN=∑i∑jp(i,j∣θ)∑j[∑ip(i,j∣θ)]2
5.Run Percentage(RP)
R P = ∑ i ∑ j p ( i , j ∣ θ ) N p RP=\sum_i\sum_j\frac{p(i,j|\theta)}{N_p} RP=i∑j∑Npp(i,j∣θ)
N p N_p Np是GLRLM中元素个数
6.Low Gray Level Run Emphasis(LGLRE)
L G L R E = ∑ i ∑ j [ p ( i , j ∣ θ ) i 2 ] ∑ i ∑ j p ( i , j ∣ θ ) LGLRE=\frac{\sum_i\sum_j[\frac{p(i,j|\theta)}{i^2}]}{\sum_i\sum_jp(i,j|\theta)} LGLRE=∑i∑jp(i,j∣θ)∑i∑j[i2p(i,j∣θ)]
7.High Gray Level Run Emphasis(HGLRE)
H G L R E = ∑ i ∑ j i 2 p ( i , j ∣ θ ) ∑ i ∑ j p ( i , j ∣ θ ) HGLRE=\frac{\sum_i\sum_ji^2p(i,j|\theta)}{\sum_i\sum_jp(i,j|\theta)} HGLRE=∑i∑jp(i,j∣θ)∑i∑ji2p(i,j∣θ)
8.Short Run Low Gray Level Emphasis(SRLGLE)
S R L G L E = ∑ i ∑ j [ p ( i , j ∣ θ ) i 2 j 2 ] ∑ i ∑ j p ( i , j ∣ θ ) SRLGLE=\frac{\sum_i\sum_j[\frac{p(i,j|\theta)}{i^2j^2}]}{\sum_i\sum_jp(i,j|\theta)} SRLGLE=∑i∑jp(i,j∣θ)∑i∑j[i2j2p(i,j∣θ)]
9.Short Run High Gray Level Emphasis(SRHGLE)
S R H G L E = ∑ i ∑ j [ i 2 p ( i , j ∣ θ ) j 2 ] ∑ i ∑ j p ( i , j ∣ θ ) SRHGLE=\frac{\sum_i\sum_j[\frac{i^2p(i,j|\theta)}{j^2}]}{\sum_i\sum_jp(i,j|\theta)} SRHGLE=∑i∑jp(i,j∣θ)∑i∑j[j2i2p(i,j∣θ)]
10.Long Run Low Gray Level Emphasis(LRLGLE)
L R L G L E = ∑ i ∑ j [ j 2 p ( i , j ∣ θ ) i 2 ] ∑ i ∑ j p ( i , j ∣ θ ) LRLGLE=\frac{\sum_i\sum_j[\frac{j^2p(i,j|\theta)}{i^2}]}{\sum_i\sum_jp(i,j|\theta)} LRLGLE=∑i∑jp(i,j∣θ)∑i∑j[i2j2p(i,j∣θ)]
11.Long Run High Gray Level Emphasis(LRHGLE)
L R H G L E = ∑ i ∑ j i 2 j 2 p ( i , j ∣ θ ) ∑ i ∑ j p ( i , j ∣ θ ) LRHGLE=\frac{\sum_i\sum_ji^2j^2p(i,j|\theta)}{\sum_i\sum_jp(i,j|\theta)} LRHGLE=∑i∑jp(i,j∣θ)∑i∑ji2j2p(i,j∣θ)
12.Gray-Level Variance (GLV)
13.Run-Length Variance (RLV)
《纹理特征一:GLCM—灰度共生矩阵》
《纹理特征三:GLSZM–灰度区域大小矩阵》
《纹理特征四:NGTDM–邻域灰度差矩阵》