1 / 67

Overview of Feature Design and Classification Result

Overview of Feature Design and Classification Result. Xiang Chen Yanhua Hu Juchang Hua Ting Zhao Sept 23, 2004. 2D Cell Level Features. 2D Features Morphological Features. Contain 22 features Describing the aspect of images what human observer may pay attention to.

denise
Download Presentation

Overview of Feature Design and Classification Result

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Overview of Feature Design and Classification Result Xiang Chen Yanhua Hu Juchang Hua Ting Zhao Sept 23, 2004

  2. 2D Cell Level Features

  3. 2D FeaturesMorphological Features • Contain 22 features • Describing the aspect of images what human observer may pay attention to

  4. 2D FeaturesMorphological Features Object features

  5. Equation - COF • Center Of Fluorescence

  6. 2D FeaturesMorphological Features Object features (DNA)

  7. (Boland and Murphy, 2001) 2D FeaturesMorphological Features # of objects Average size of objects Average distance to COF 108 83 31 6 232 4

  8. 2D FeaturesMorphological Features Edge features

  9. Canny method of edge detection Convex Hull finding Illustration – Edge Detection and Convex Hull

  10. 2D FeaturesMorphological Features Skeleton features

  11. Skeleton finding Illustration – Skeleton

  12. Zernike Moment Features(SLF 3.17-3.65) • Shape similarity of protein image to Zernike polynomials Z(n,l) • 49 polynomials and 49 features left: Zernike polynomials A: Z(2,0) B: Z(4,4) C: Z(10,6) right: lamp2 image

  13. Haralick Texture Features(SLF7.66-7.78) • Correlations of adjacent pixels in gray level images • Co-occurrence matrix P: N by N matrix, N=number of gray level. Element P(i,j) is the probability of pixels with value i being adjacent with pixels with value j • 13 statistical features

  14. 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 0 2 1 3 1 2 1 0 1 1 0 1 0 3 1 0 3 0 1 2 2 4 4 4 2 1 6 3 4 2 1 4 3 3 2 3 0 4 4 3 1 4 2 2 3 0 3 6 2 3 0 3 4 1 3 0 4 0 3 4 2 3 2 2 4 1 4 2 4 4 3 3 1 2 4 1 4 3 2 Co-occurrence Matrix

  15. Wavelet Transformation A: approximation (low frequency) D: detail (high frequency) X=A3+D3+D2+D1

  16. Daubechies D4 Wavelet The scaling function (for low frequency component) The mother wavelet (for high frequency component)

  17. Daubechies D4 decompsotion Original image Wavelet Transformation

  18. Feature Calculation • Preprocessing • Background subtraction and thresholding, • Translation and rotation • Wavelet transformation • The Daubechies 4 wavelet • 10th level decomposition • The average energy of the three high-frequency components

  19. A 2D Gabor Function A 2D Gaussian modulated by a sinusoid

  20. Graph for Gabor Function We can extend the function to generate Gabor filters by rotating and dilating

  21. Feature Calculation • Preprocessing • 30 Gabor filters were generated using five different scales and six different orientations • Convolve an input image with a Gabor filter • Take the mean and standard deviation of the convolved image • 60 Gabor texture features

  22. Classification Result based on 2D Cell Level Features

  23. Typical 2D images

  24. Different Classifiers • Decision Tree • K-NN • Neural Net • Single hidden layer • Multiple hidden layers • SVM • Linear kernel • RBF kernel • Exponential RBF kernel • Polynomial kernel • AdaBoost • Bagging • Mixture of Experts • Majority-voting Ensemble

  25. Comparison of Different 2D Classifiers

  26. Majority Voting Ensemble Classifier with More Features • Daubechies wavelet and Gabor transform features • SLF 15 :44 features selected by SDA from full feature set except DNA features • SLF 16: 47 features selected by SDA from full feature set including DNA features

  27. Confusion Matrix using SLF16

  28. 2D Field Level Features

  29. 2D FeaturesField Features • Contain 26 features • These features are not sensitive to the number of cells in the field of image • Derived from the 2D morphological features (3 obj. + 5 edge + 5 skeleton) and Haralick texture feature (13)

  30. 2D FeaturesField Features

  31. 2D FeaturesField Features (con’t)

  32. 2D Field Level Classification Result

  33. Field Feature ClassificationData - 10 Class mHela Images (Kai and Murphy, 2004) • Has various number (2-6) of cells • Created by randomly mixing the cropped single Hela cell images

  34. Field Feature ClassificationMethod • SDA feature selection • 23 selected • Top 7 are Haralick texture features • Classier • DAG Gaussian-kernel SVM • Max-win multi-class strategy

  35. Field Feature ClassificationResult (Kai and Murphy, 2004)

  36. 3D Cell Level Features

  37. SLF-9 • 28 features, 14 from protein objects and 14 from their relationship to corresponding DNA images • Based on number of objects, object size, object distance to COF • Corresponding DNA image required • A subset of 9 features selected by SDA forms SLF10

  38. SLF-14 • 14 SLF-9 features that do not require DNA images • 2 Edge features • Ratio of above threshold pixel along an edge • Ratio of fluorescence along an edge • 26 3D Haralick texture feature • GLCM built on 13 directions • One set (13) of mean features and the other set (13) of range features

  39. Pixel Resolution and Gray Levels • Texture features are potentially influenced by the number of gray levels and pixel resolution of the image • Optimization for each image dataset required

  40. SLF-17 • A feature subset with 7 features selected from SLF-14 at 256 gray levels and 0.4 micron pixel resolution • 1 morphological feature • 1 edge feature • 5 texture features • Achieved 98% overall accuracy in a 10-class 3D HeLa dataset

  41. Classification Result Based on 3D Cell Level Features

  42. Typical 3D images

  43. Classification using Different Feature Set • SLF-9: 91% • SLF-10: 93% • 14 morphological features from SLF-9: 86% • SLF-17: 98%

  44. Net Benefit of 3D Texture Features • Consistently better performance using 256 gray levels compared to the other two gray levels • Comparable performance using 0.2 μm and 0.4 μm pixel resolutions

  45. Confusion Matrix using 3D-SLF17

  46. 2D/3D Object Level Features and Classification Result

  47. 2D Object Features SOF1.1: Number of pixels in object SOF1.2: Distance between object COF and DNA COF SOF1.3: Fraction of object pixels overlapping with DNA SOF1.4: A measure of eccentricity of the object SOF1.5: Euler number of the object SOF1.6: A measure of roundness of the object SOF1.7: The length of the object’s skeleton by homotopic thinning SOF1.8: The ratio of skeleton length to the area of the convex hull of the skeleton SOF1.9:The fraction of object pixels contained within the skeleton SOF1.11: The fraction of object fluorescence contained within the skeleton SOF1.12: The ratio of the number of branch points in skeleton to length of skeleton

  48. Feature Description • SOF1.4 A measure of eccentricity of the object • To measure the eccentricity of the ellipse that is equalivent

  49. Feature Description • SOF1.6: A measure of roundness of the object

More Related