Image and Video Analysis
Tracking objects in videos is a challenging task, due to varying light conditions, camera angle, and other environmental disturbing factor, as well as occlusion, both partial and total. In  the authors propose a novel pairwise diversity measure, that recalls the Fisher linear discriminant, to construct a classifier ensemble for tracking a non-rigid object in a complex environment. A subset of constantly updated classifiers is selected exploiting their capability to distinguish the target from the background and, at the same time, promoting independent errors. Target selection and tracking can also be assimilated to this first type of problems. Multiple Description Coding is used in . The proposal was inspired from the framework of transmitting data over heterogeneous network, especially wireless network.
Due to the increasing amount of media produced, stored, retrieved over the Internet, video and image annotation and segmentation are becoming more and more important, allowing to index and categorize huge amounts of media. In  images are annotated with a description of the content in order to facilitate the organization, storage and retrieval of image databases. Several features have been designed and experimentally compared, producing a classifier that can provide a reasonably good performance on a generic photograph database. Movie segmentation is discussed instead in  and , where combinations of audio and visual features are exploited for effective scene cut and scene type recognition. Image segmentation is also performed, in , by using multi-scale features for pixel classification. The link between scale selection and the maximum combination rule from pattern recognition is explored. Automatic annotation is investigated in . Such paper deals, in particular, with the annotation of sports videos. The concept of “cues” allows to attach semantic meaning to low-level features computed on the video and audio. Shots are classified, based on the cues they contain, into the sports they belong to. Diversity is tackled by using the RGB color space, which is prominent both as a color scheme, and a display scheme. The authors of , too, claim that the recent advent of multiple classifier systems provides the unique opportunity to exploit the diverse information encapsulated in the different color representations in a systematic fashion. They use information gathered in different color spaces, and subsequently use suitable measures to investigate the diversity of the information infused by the different color spaces.