ZW quasi-patent • High-fidelity Two-value graph detail segmentation algorithm

ZW quasi-patent • High-fidelity Two-value graph detail segmentation algorithm
The detailed segmentation algorithm of the high-fidelity binary graph is a derivative work of the Chinese Font Association project.
Said quasi-patented algorithm, is because of the image algorithm standards do not understand, although reported patents, but not in this respect, need to consult a professional patent consultant.
Prototype is implemented with Opencv+python, because Halcon, for the Association of Designers, the threshold is too high, so deliberately designed a set of Opencv+python LIVE-CD, decompression can, without configuration.

Chinese traditional calligraphy, there are many feibai, ink-based techniques, produced a lot of small holes, solitary point, from the perspective of the image, these are small infected areas.

Traditional image processing algorithms, such as license plate, face, industrial monitoring, etc., can be solved simply by iterative filtering.
But for calligraphic works, these details are an integral part of the work.
Compared with industrial image processing, the images of calligraphy fonts are all two value graphs,
The surface looks like these image data sources are very clean, but because of the lack of detail, especially in the stroke decomposition, very troublesome.

Calligraphy works, different styles, changes, can not be modeled, template matching means of segmentation, the cost of the construction of the library is too high.

(GB level two, each set of fonts, are included in 6,763 characters,)

Through the multi-party search, there is no ready-made algorithm, can refer to, all need to start from scratch, self-development.
Through repeated groping, debugging, the author successfully through the image morphology, combined with contour matching, the perfect realization of the relevant functions, by the relevant scholars recognized.

Figure 1, is the original, Figure 2, is the effect of the segmentation

Note The Strokes at Fig. Two, which have been split, B, C two, the splash effect below point A, and other details, are completely preserved.

A, is the most difficult to cut, but also keep the details of the area, the rest of the joint, the cutting is very simple, it is not introduced here.

This cutting algorithm is another difficulty is that the national standard two, each set of fonts, are included in 6,763 characters,

The algorithm must be universal, all glyphs of each set of fonts, and various other fonts, all applicable.

Otherwise, separate for each word to split, a single set of development, implementation costs, it is unacceptable, let alone many sets of fonts,

In addition, the appendix of a few pictures, is the other processing links, for reference.

ZW quasi-patent • High-fidelity Two-value graph detail segmentation algorithm