Image Features Correlation with the Impression Curve for Automatic Evaluation of the Computer Game Level Design

Abstract

In this study, we present the confirmation of existence of the correlation of the image features with the computer game level Impression Curve. Even a single image feature can describe the impression value with good precision (significant strong relationship, Pearson r> 0,5). Best results were obtained using by combining several image features using multiple regression (significant very strong positive relationship, Pearson r = 0,82 at best). We also analyze the different set of image features at different level design stages (from blockout to final design) where significant correlation (strong to very strong) was observed regardless of the level design variant. Thanks to the study results, the user impression of virtual 3D space, can be estimated with a high degree of certainty by automatic evaluation using image analysis.