分类器特征提取

Halcom

对应一组图像而言，图像分类应用较为广泛，如何选取特征很关键：

1. FeatureVector := []
   
2. * Compute features.
   
3. gen_sobel_features (Image, FeatureVector, FeatureVector)
   
4. * Downscale the image (image pyramid) and compute features.
   
5. zoom_image_factor (Image, Zoomed1, 0.5, 0.5, 'constant')
   
6. gen_sobel_features (Zoomed1, FeatureVector, FeatureVector)
   
7. * Uncomment lines to use further pyramid levels:
   
8. zoom_image_factor (Zoomed1, Zoomed2, 0.5, 0.5, 'constant')
   
9. * gen_sobel_features (Zoomed2, FeatureVector, FeatureVector)
   
10. * zoom_image_factor (Zoomed2, Zoomed3, 0.5, 0.5, 'constant')
    
11. * gen_sobel_features (Zoomed3, FeatureVector, FeatureVector)
    
12. * zoom_image_factor (Zoomed3, Zoomed4, 0.5, 0.5, 'constant')
    
13. * gen_sobel_features (Zoomed4, FeatureVector, FeatureVector)
    
14. FeatureVector := real(FeatureVector)

复制代码

gen_sobel_features如下：

1. * Coocurrence matrix for 0 deg and 90 deg:
   
2. cooc_feature_image (Image, Image, 6, 0, Energy, Correlation, Homogeneity, Contrast)
   
3. cooc_feature_image (Image, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
   
4. * Absolute histogram of edge amplitudes:
   
5. sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
   
6. gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
   
7. * Entropy and anisotropy:
   
8. * entropy_gray (Image, Image, Entropy, Anisotropy)
   
9. * Absolute histogram of gray values:
   
10. * gray_histo_abs (Image, Image, 8, AbsoluteHistoImage)
    
11. * Add features to feature vector:
    
12. FeaturesExtended := [Features,Energy,Correlation,Homogeneity,Contrast]
    
13. FeaturesExtended := [FeaturesExtended,AbsoluteHistoEdgeAmplitude]
    
14. * FeaturesExtended := [FeaturesExtended,Entropy,Anisotropy]
    
15. * FeaturesExtended := [FeaturesExtended,AbsoluteHistoImage]

复制代码

金字塔式的纹理特征求解；

qianlongzaiyuan

太棒了，我想尽快升级会员学习

LJN

学习啦，楼主的贴子很有用

liuxingan

太棒了，我想尽快升级会员学习