The most important presentation methods for key-frame summaries were also discussed.
However this method is not based on shot identification, since it produces 3D key-frame summaries without requiring prior video shot detection.
The quality of key-frame summaries obtained using these approaches can be negatively affected when accurate shot segmentation is not available.
The experiments can include a set of absolute evaluations and/or a set of relative evaluations, in which two key-frame summaries are presented and compared.
One other important and interesting research topic is the design and implementation of methods and tools to present 3D key-frame summaries.
Unfortunately for the case of 3D video, there is not as yet any similar framework, where key-frame summaries and the respective original sequences are available for research use.
However, the result description method has some limitations, such as the reduced number of methods which can be compared at same time, i.e. it is inadequate to compare key-frame summaries of a large number of video sequences or methods.
Another challenging topic in the research of 3D key-frame summarisation is the design of an efficient and intuitive visualisation interface that allows easy navigation and visualisation of the key-frame summaries.
In Liu et al. [66], the similarity measure chosen was peak signal-to-noise ratio (PSNR), but other similarity metrics that include 3D features can also be used in the evaluation of 3D key-frame summaries.
More recently, Avila et al. [24] also proposed another evaluation setup, wherein the original video and the key-frame summaries of several methods are available for downloading, together with the results of several key-frame extraction methods for 2D video.
During the evaluation of the key-frame summaries, it is also required taking into account the external factors which can influence the ratings of the summaries, such as the attention and fatigue specially when there are long evaluation sessions with many video summaries.
