Haralick features

Haralick features provide information on how the intensities of pixels in images of a certain position are related with the intensities of neighboring pixels. These features originated from the co-occurrence matrix. In the Cellprofiler  used to measure the local features, there is a method called MeasureTexture. This method can obtain textures of images at different scales. The scale can be chosen by the user. The co-occurrence matrix thus constructed is a result of the scale chosen by the user. If we chose a scale of 2, then each pixel from the image is compared to the pixel that is present at two pixels at the right side of the given pixel. This method usually quantizes the given image into 8×8 co-occurrence matrix. It looks for the number of pixels and neighbors each having the combinations of 8×8 intensity. Using this method, around 13 features are calculated. These features are calculated on the co-occurrence matrix by making certain calculations. The features calculated using this method are angular second moment, correlation, contrast, variation (sum of squares), inverse difference moment, sum variance, sum average, entropy, sum entropy, difference entropy, difference variance, information measure of correlation1 and correlation2.

Gabor wavelet features

The Gabor wavelet features are the result of applying Gabor filters to images. These features are similar to wavelet features. The frequency measure in different orientations of images is measured using the Gabor features. Though they are similar to and work like wavelets in this context, these Gabor features are not wavelets by the strict definition of mathematics. These features detect bands of intensities that are correlated.


The module is used generally to try series of structure elements that outputs a number of measures. It defines whether the structure elements fit in the texture of given images.