Improvement of the Histogram for Infrequent Color-Based Illustration Image Classification

Illustration images have a style that is often characterized by its emotional features. In this research, we tried to classify an illustration image based on the style. We hypothesized that the color that low appeared frequently would help to analyze the style. In previous research, an Infrequency histogram (IF-hist), was proposed as a feature that focused on the color appearance infrequency. However, IF-hist has the weakness that this feature cannot be used effectively for certain kind of images. To address this issue, we proposed a color histogram which uses a histogram value as its threshold. To evaluate the effectiveness of the proposed method, we experimented with classifying the illustration image into the style of “For boys” and ```For girls". As a result of the experiment, when using the proposed histogram as the feature, classification performance was better than using a normal color histogram. Moreover, the F-measure of the proposed histogram was better than the F-measure of IF-hist.