Skip to main content

Table 1 Composition of the CHARM-inspired CellProfiler feature vector. The groups and levels construction of CHARM was recreated, although the final measurement set is not identical

From: CP-CHARM: segmentation-free image classification made accessible

High-contrast features

Polynomial decompositions

Pixel statistics

Textures

Edge

Gabor

Image

Chebyshev

Chebyshev-

Moments ∗ (4)

Multiscale

Haralick

Tamura

statistics (4)

features ∗ (2)

statistics (15)

statistics (32)

Fourier

 

histogram ∗ (24)

textures ∗ (104)

textures ∗ (6)

    

Statistics (32)

    

Mean

Gabor features computed at four angles

Max ∗

32-bins histogram of the 400 coefficients of the Chebyshev transform of the image

Modulus of the complex coefficients of the Fourier transform of the Chebyshev transform of the image

Mean (1st)

3-bins histogram

Statistics based on the co-occurence matrix of the image

Contrast

Max

 

Mean ∗

  

Variance (2nd)

  

Coarseness

Variance

 

Percent minimal ∗

   

5-bins histogram

 

Directionality

Number of edge pixels

    

Skewness (3rd)

  

3-bins histogram of coarseness

  

Percent maximal ∗

   

7-bins histogram

  
     

Kurtosis (4th)

   
  

Variance ∗

   

9-bins histogram

  
  

Total intensity ∗

      
  

Mean intensity after thresholding

      
  

Variance on thresholded image

      
  

Number of pixels above threshold

      
  1. *denotes features extracted on higher image levels, namely on the original image, on its Wavelet transform, on its Chebyshev transform, on its Fourier transform, on the Wavelet transform of its Fourier transform, and on the Chebyshev transform of its Fourier transform