How KanjiTomo works?

This page contains a description of KanjiTomo's OCR algorithm.

The original image Image is sharpened Gray pixels are removed
Touching pixels are combined into areas Each area represents a fragment of character Fragments are grouped into characters
Target character is aligned with reference characters.
Alignments are scored by counting common pixels. Pixels are represented as 32x32 bitmaps and comparisons are done with fast bitwise operations.
This way all common Japanese characters can be compared against each target character. Various transformations (such as stretching and offset)
are also done to find best possible alignment.