Graininess Detection

(I) What is graininess?
  • Graininess corresponds to high frequency content in the picture

(II) How do we measure graininess?
  • If we low pass filter an image, then the graininess in the picture will become blurry


Monet after filtering


Rembrandt after filtering

(III) How exactly do we low pass filter?
  • Apply a Wiener filter, which is designed to filter out constant power additive noise (MATLAB help   documentation).
• The Wiener filter adaptively filters using statistics estimated from a local neighborhood of each pixel (MATLAB help documentation).
• The size of the neighborhood is set to be one-tenth of the size of the original picture.

(IV) How do we measure blurriness?
  • Blurriness corresponds to the amount of contrast in the picture.

High Contrast

Low Contrast

(V) How do we measure contrast?
  • Convert the image to gray scale in MATLAB.
• The image is then a matrix of values from zero to one where each value corresponds to the darkness of a pixel.

Example: Image of size 3x4 pixels

  • Approximate the contrast by finding the sum of the differences between elements of a row and the sum of the differences of elements of a column.

Row contrast of example = 3.4
Column contrast of example = 2.7
  • Add two sums for total contrast.

Total contrast of Example= Row contrast + Column Contrast = 6.1

(VI) How do we compare the graininess of different pictures of different sizes?
  • Measure the percent change in contrast to determine the blurriness.

(Contrast of Original Contrast of Blurred)/ Contrast of Original

(VII) Summary of Algorithm to Measure Graininess

(VIII) Graininess Trends Found Among Three Artists

Graininess Trends: Approximated by Measuring Blurriness After Wiener Low Pass Filter
  • Monet paintings found in percentage range of 76 to 92.
• Rembrandt paintings have no definite trend.
• Picasso paintings found in percentage range of 67 to 85.
• Results interesting but not useful for classification since ranges of artist overlap.

  Graininess Analysis MATLAB code