Automatic Evaluation and Generation of
Aesthetic Chinese Calligraphy



Back to Homepage

I. Calligraphic Characters Representation and Its Acquisition


The input character Its skeletonization
result
Comparing with its
standard font
the stroke extraction
result on skeleton
the stroke extraction
result

An example of stroke extraction from an image of a calligraphic character

The original character Its skeleton trajectory With covering ellipses
every 50 pixels
With covering ellipses
every 20 pixels
With covering ellipses
every 5 pixels

An example of calligraphy character representation

Back to Top    Back to Homepage

II. Evaluating Shapes of Individual Strokes


The strokes signals used for shape grading: w = {Sx ,Sy ,Ma,Mi,D}

(a) the evaluation results for individual strokes

(b) the evaluation given by people

Examples of individual strokes evaluation
(A completely black stroke indicates it is "good" while a completely red one means "bad")

Back to Top    Back to Homepage

III. Evaluating the Spatial Layout between Strokes


100.%-0.00%-0.00% 61.6%-38.4%-0.00% 86.8%-13.2%-0.00% 35.5%-60.0%-4.50% 31.3%-60.0%-8.70% 55.0%-45.0%-0.00% 0.00%-54.3%-45.7%
100.%-0.00%-0.00% 70.0%-30.0%-0.00% 100.%-0.00%-0.00% 40.0%-60.0%-0.00% 70.0%-30.0%-0.00% 90.0%-10.0%-0.00% 50.0%-50.0%-0.00%
0.00%-54.3%-45.7% 0.00%-50.7%-49.3% 98.4%-1.60%-0.00% 100.%-0.00%-0.00% 50.3%-49.7%-0.00% 94.0%-6.00%-0.00% 57.6%-42.4%-0.00%
20.0%-60.0%-20.0% 20.0%-60.0%-20.0% 100.%-0.00%-0.00% 60.0%-40.0%-0.00% 70.0%-30.0%-0.00% 80.0%-20.0%-0.00% 70.0%-30.0%-0.00%
90.2%-9.80%-0.00% 48.1%-51.9%-0.00% 0.00%-0.00%-100.% 0.00%-0.00%-100.% 0.00%-0.00%-100.% 0.00%-0.00%-100.% 0.00%-5.90%-94.1%
80.0%-20.0%-0.00% 50.0%-50.0%-0.00% 0.00%-40.0%-60.0% 0.00%-0.00%-100.% 0.00%-0.00%-100.% 0.00%-0.00%-100.% 0.00%-0.00%-100.%

(The percents in each unit rate the visual quality of that character, where three values represent its probabilities to be "good", "so-so" and "bad" respectively from left to right. For comparison, the bottom one comes from humans and the upper one comes from our algorithm.)
Back to Top    Back to Homepage

IV. Evaluating Coherence of Stroke Styles


100.%-0.00%-0.00% 3.90%-61.2%-34.9% 91.0%-9.00%-0.00% 62.7%-37.3%-0.00% 0.00%-0.00%-100.%
100.%-0.00%-0.00% 20.0%-60.0%-20.0% 70.0%-30.0%-0.00% 100.%-0.00%-0.00% 0.00%-20.0%-80.0%

100.%-0.00%-0.00% 0.00%-9.30%-90.7% 0.00%-12.0%-88.0% 58.6%-41.4%-0.00%
100.%-0.00%-0.00% 0.00%-40.0%-60.0% 0.00%-50.0%-50.0% 40.0%-60.0%-0.00%

(The percents in each unit rate the visual quality of that character, where three values represent its probabilities to be "good", "so-so" and "bad" respectively from left to right. For comparison, the bottom one comes from humans and the upper one comes from our algorithm.)
Back to Top    Back to Homepage

V. The Overall Evaluation


100.%-0.00%-0.00% 64.5%-35.5%-0.00% 55.7%-44.3%-0.00% 67.8%-32.2%-0.00% 71.1%-28.9%-0.00% 0.00%-16.4%-83.6% 100.%-0.00%-0.00%
100.%-0.00%-0.00% 60.0%-40.0%-0.00% 60.0%-40.0%-0.00% 70.0%-30.0%-0.00% 80.0%-20.0%-0.00% 0.00%-0.00%-100.% 100.%-0.00%-0.00%

43.8%-56.2%-0.00% 53.7%-46.3%-0.00% 24.4%-67.3%-8.30% 14.2%-57.5%-28.3% 0.00%-57.0%-43.0% 100.%-0.00%-0.00% 44.2%-55.8%-0.00%
20.0%-80.0%-0.00% 10.0%-90.0%-0.00% 30.0%-70.0%-0.00% 20.0%-60.0%-20.0% 0.00%-10.0%-90.0% 100.%-0.00%-0.00% 30.0%-60.0%-10.0%

(The percents in each unit rate the visual quality of that character, where three values represent its probabilities to be "good", "so-so" and "bad" respectively from left to right. For comparison, the bottom one comes from humans and the upper one comes from our algorithm.)
Back to Top    Back to Homepage

VI. Automatic Calligraphy Generation


(The generated calligraphy fonts are improved with the help of grading module. In each row, the evaluating grades increase from left to right, corresponding to incremental improvements on the appearance.)

(Automatically generated calligraphy of an ancient poetic verse from the Tang Dynasty in the 8th century. The five learning sources in this experiment are shown in (u-1)每(u-5). Ten automatic calligraphy generation results using the calligraphy visual quality grading component as the feedback are shown in (u-6)每(u-15). For comparison, we show five automatic calligraphy generation results without employing the grading component as the feedback in (u-16)每(u-20). As judged by many Chinese scholars participating in the experiment, the visual quality of the calligraphic writings in (u-6)每(u-15) is clearly and significantly better than that of (u-16)每(u-20).)
Back to Top    Back to Homepage