User Study

User Study for Structure-based ASCII Art

We conducted a user study to compare our results to those manually created by artists and those generated by the method of O'Grady and Rickard. We evaluate (1) the similarity between the input and ASCII art, and (2) the clarity of content presented in the ASCII art without referring to the input. Three test sets are prepared, and 30 participants were invited to grade the results in a 9-point scale ([1-9] with 9 being the best).

The following shows the three test sets, together with the statistics collected. In each test set, images (a), (b) and (c) are created by an artist, our method, and the method of O'Grady and Rickard, respectively. Each cell in the following tables is listed in the form of value1/value2, with value1 denotes similarity, and value2 denotes the clarity.

In all tests, our method is comparable to artists manual creation, while the method of O'Grady and Rickard is the worst. Looking into the statistics of our method and the artist creation, one can find that our method typically has a higher similarity than clarity, while artist creation has a higher clarity than similarity. Because artists can intelligently modify the content while our method only straightly follows in the input content.

Test set 1:

(Similarity/Clarity)	Mean	Standard deviation	95% confidence interval
(Similarity/Clarity)	Mean	Standard deviation	Lower Bound	Upper Bound
Artists	7.19/7.51	1.26/1.14	6.77/7.12	7.61/7.90
Our method	6.74/6.51	1.19/1.20	6.34/6.11	7.13/6.92
O'Grady and Rickard	4.12/3.85	1.66/1.82	3.56/3.24	4.68/4.47