Adaptive Image Binarization

The adaptive binarization is based on second moments: Given the value of the center pixel within the sliding window, the output pixel is classified as either black or white depending on whether the second moment to the "left" of this value is smaller or larger than the second moment to the "right" ("to the left" means "for the part of the histogram with pixel values smaller than the center pixel value", and correspondingly for "Right") However, if the total sum of second moments within the sliding window is smaller than a certain contrast, the area is considered to contain only one pixel class, and the output pixel is set to binary.

Sliding window size: Size (in pixels) of square sliding window. This determines the local area of the image used to calculate each intensity histogram distribution.

Percent contrast: If the contrast between the pixels within the sliding window is smaller than the Percentage contrast then the window is considered to be uniform. Contrast is normalized to 100.

set = 100 for a reasonable result

set = 0 to force all regions to be accepted as non-homogeneous

set = 10000 (or higher) to only binarize very high contrast regions

Sample cases:

Figure 1 Original image (left) and same image after binarization with Window size= 30 (right) and Percent contrast=100