Conciseness is related to the length of the key-summary, while the coverage evaluation is based on comparison between computed key-frames summary and ground-truth summary, expressed by the recall rate, precision rate, CUS A and CUS E. Most evaluation metrics reviewed above were developed for 2D video.
For this reason it is important to quantify the conciseness of the key-frame summary.
In this section, we present current key-frame summary evaluation methods and some related issues.
Conciseness, coverage, context and coherence are desirable attributes in any key-frame summary.
To judge the conciseness of a key-frame summary a measure of the Compression Ratio (CR) is used [68].
Thus, the size of key-frame summary corresponds to the number of video shots of the entire video.
The key-frame presentation methods are independent of the key-frame extraction operation and thus the same key-frame summary can be presented to viewers in different ways.
A video skim can be generated from a key-frame summary by adding frames or segments that include the key-frames, while a key-frame summary can be created from a video skim by uniform sampling or selecting the most representative frame from each video skim segment [4].
A K key-frame summary is a good representation of the original F when the magnitude of its SRD is high.
The key-frame summary sought should minimise a Conciseness cost function, which is a weighted sum of the Rate and Distortion functions defined in the work.
In this type evaluation, a panel of viewers are asked to observe both the summaries and the original sequence and then respond to questions related to some evaluation criteria, (e.g. 'Was the key-frame summary useful?', 'Was the key-frame summary coherent?') or if each key-frame is 'good', 'fair', or 'poor' according to the original video sequence.
