AVSA: An automatic video segmentation application

• Tiago H. Trojahn USP
• Rudinei Goularte USP

Resumo

Segmentation is an important preprocessing step for a number of current multimedia applications using video, like recommendation, personalization or indexing. Since manual segmentation is prone to interpretation error and are time demanding, researchers concentrate efforts in developing automatic segmentation methods. Automatic techniques need a number of technical input parameters, requiring specialists to be operated. Moreover, these techniques need the specialist to give a threshold, used to decide when a shot or scene transition occurs. Obtaining an adequate threshold is time consuming and mostly an empirical process. The precision is greatly affected by particularities of the input video, so, an inadequate threshold can lead to over or under-segmentation. To addresses these problems, this paper presents a friendly user application developed in Java which has two main contributions: perform video segmentation using both an automatic method for calculate the needed threshold and a heuristic to overcome some gradual shot transitions issues. The application, named AVSA, uses the video histogram intersection or histogram absolute differences to perform the segmentation. Furthermore, performance tests are presented in order to testify the precision and the recall of the application when segmenting newscast videos.