1. Derivative Based Algorithms:
- Thresholded absolute gradient:
wheref=∑x∑y|I(x+1,y)−I(x,y)| |I(x+1,y)−I(x,y)|>threshold; - Squared gradient:
wheref=∑x∑y(I(x+1,y)−I(x,y))2 (I(x+1,y)−I(x,y))2>threshold; - Brenner gradient:
wheref=∑x∑y(I(x+2,y)−I(x,y))2 (I(x+2,y)−I(x,y))2>threshold; - Tenenbaum gradient:
f=∑x∑y(SobelX2(x,y)+SobelY2(x,y)); - Energy Laplace:
wheref=∑x∑y((L∗I)(x,y))2 L=[−1−4−1−420−4−1−4−1] - Sum of modified Laplace:
f=∑x∑y(LaplaceX(x,y)|+|LaplaceY(x,y)|) - Sum of squared Gaussian derivatives
wheref=1total pixels∑x∑y((Gaussian derivative X(x,y))2+(Gaussian derivative Y(x,y))2) ,σ=d/(2√3) = dimension of the smallest feature;d
2. Statistical Algorithms:
- Variance Measure ;
wheref=1total pixels∑x∑y(I(x,y)−―I)2; is the mean of image.―I - Normalized Variance Measure ;
f=1total pixels×―I∑x∑y(I(x,y)−―I)2; - Auto-correlation Measure:
f=∑x∑yI(x,y)I(x+1,y)−∑x∑yI(x,y)I(x+2,y); - Standard Deviation-based Correlation:
f=∑x∑yI(x,y)I(x+1,y)−―I2(x,y)×total pixels;
3. Histogram-based Algorithms:
- Range Algorithm:
f=max{i|histogram(i)>0}−min{i|histogram(i)>0}; - Entropy Algorithm:
wheref=−255∑i=0p(i)log2p(i) p(i)=histogram(i)/total pixels;
4. Other Algorithms:
- Threshold Contents:
wheref=∑x∑yI(x,y) I(x,y)≥threshold; - Thresholded Pixel Count:
wheref=∑x∑yθ(threshold−I(x,y)); is the step-function.θ(x) - Image Power:
wheref=∑x∑yI2(x,y) I(x,y)≥threshold;
Ref: Dynamic evaluation of autofocusing for automated microscopic analysis of blood smear and pap smear, J. Microscopy,227, 15(2007).
728x90
'Image Recognition' 카테고리의 다른 글
Union-Find Connected Component Labeling (0) | 2012.11.01 |
---|---|
RANSAC: Ellipse Fitting (1) | 2012.10.07 |
Statistical Region Merging (2) | 2012.03.25 |
Local Histogram Equalization (0) | 2012.03.10 |
2차원 Savitzky-Golay Filters 응용 (0) | 2012.02.28 |