Video Digesting Technique for Baseball Games Based on Information Entropy

We describe an automatic video digesting technique for broadcasted baseball games based on information entropies derived from an overlayed score ticker. The overlayed score ticker of baseball games displays the current score of each team, the inning, and the number of balls/strikes/outs, and whether or not there are runners on base. This information is used to estimate specific events in baseball that we defined: 3 runner situations, 8 batting results, and 4 types of scores. Rare events such as a “home run” have higher information entropy, and frequent events such as “out” have lower entropy. We suppose that an event with higher information entropy is a more important scene for video digesting. Thus, the scenes with higher entropies are selected by priority and embedded into a limited timeline. In our experiment, we applied our technique to broadcasted baseball games and compared it with a conventional sound volume based technique. The two techniques were evaluated subjectively by watching generated videos. We found that our technique has higher performance.