Conclusion

We described the original algorithm initially used in our system for video segmentation, indexing and generating key frames within a video segment. Our approach partitions the selected frames into exclusive blocks in order to resolve the problem of missed cuts. The system uses initially spatial and temporal skip parameters to improve the performance of the analysis.

However, to compensate for the use of the temporal skip, the original algorithm re-analyzes all frames in the region if the first test shows the potential to find a cut. We then explained the updated "binary penetration" algorithm. The algorithm, based on a dichotomy ad hoc image processing, makes use of the temporal correlation heuristics of the visual information within a video stream. The new approach resulted in superior performance over the original algorithm while accuracy was not sacrificed in the analysis.

We presented an extendable system that we developed for the purpose of video browsing and for detecting camera cuts to represent the visual key frames. The architecture, the functions and the uses of each module within the system were explained, together with the scenarios of data and information flow within the modules. We implemented the use of the video key framing system as a service using any java-enabled web browser. In our next steps, we will augment the prototype to support low capacity and processing power mobile devices such as PDA and cellular phones.