Pt1420 Unit 3.4 Glcm
3.3 Fractal dimension:
Hausdorff’s dimension is the generally used represention for fractal dimension .Considering an object that possesses an Euclidean dimension R,the Hausdorff’s fractal dimension F0 can be computed by the following expression: F0 = lim┬(e→0)log〖N(e〗 )/log〖e⁻¹〗 where N(e) is the counting of hyper-cubes of dimension R and length e that fill the object.
But here fractal dimension is obtained using box counting algorithm .[15]
3.4 GLCM is the widely used statistical method for feature extraction.The number of gray levels present in the input image becomes the number of rows and columns in the matrix.
In a GLCM matrix, N is the number of gray levels in an image with different values. Each element p(i,j,d,ө)
in GLCM specifies the number of times that the pixel with value i at location (x,y) occurred adjacent to a pixel with value j at a distance d and an orientation angle of ө from location (x,y).[16]
Rotation forest:
Rotation forest is an ensemble method. Ensemble model provides more accurate classification compared to individual based classifier. Rotation forest is based on trees,thus has the name’forest’.[18]
It divides the obtained feature set data into k random non-overlapping subsets.Principle component analysis is applied to each of the subsets.Principle Component Analysis is an unsupervised approach where the number of variables is decreased and relatively important components are retained.These components are called principle components and they represent the best amount of variance from the set. Rotation matrix is formed with the features as the columns and the principle components obtained arranged corresponding to the columns in row wise manner.A decision tree is formed using the following rotation matrix.Such collection of decision trees forms the rotation forest.[19]
The advantage of rotation forest is said to show good results even with less number of trees.[20]
Hausdorff’s dimension is the generally used represention for fractal dimension .Considering an object that possesses an Euclidean dimension R,the Hausdorff’s fractal dimension F0 can be computed by the following expression: F0 = lim┬(e→0)log〖N(e〗 )/log〖e⁻¹〗 where N(e) is the counting of hyper-cubes of dimension R and length e that fill the object.
But here fractal dimension is obtained using box counting algorithm .[15]
3.4 GLCM is the widely used statistical method for feature extraction.The number of gray levels present in the input image becomes the number of rows and columns in the matrix.
In a GLCM matrix, N is the number of gray levels in an image with different values. Each element p(i,j,d,ө)
in GLCM specifies the number of times that the pixel with value i at location (x,y) occurred adjacent to a pixel with value j at a distance d and an orientation angle of ө from location (x,y).[16]
Rotation forest:
Rotation forest is an ensemble method. Ensemble model provides more accurate classification compared to individual based classifier. Rotation forest is based on trees,thus has the name’forest’.[18]
It divides the obtained feature set data into k random non-overlapping subsets.Principle component analysis is applied to each of the subsets.Principle Component Analysis is an unsupervised approach where the number of variables is decreased and relatively important components are retained.These components are called principle components and they represent the best amount of variance from the set. Rotation matrix is formed with the features as the columns and the principle components obtained arranged corresponding to the columns in row wise manner.A decision tree is formed using the following rotation matrix.Such collection of decision trees forms the rotation forest.[19]
The advantage of rotation forest is said to show good results even with less number of trees.[20]