Evaluation method of text detection algorithm rating
- Aliona Dangla
Several methods to evaluate text detection algorithms exist. However, they give different results, so it is difficult to estimate their relevance. In order to evaluate these methods, the proposed solution is to compare the automatic evaluations with the human evaluation by considering the latter as a reference. In order to obtain this reference, a website has been created so that a large number of users can easily classify the results of a large number of text detection algorithms on large image bases. The user interface has been developed to minimize the number of comparisons that the user must make in order to obtain a consistent ranking. The website will be tested with the results of the 2013 ICDAR competition, which is composed of height methods on a data set of 233 images.