Keyframe Extraction Algorithm for Continuous Sign-Language Videos Using Angular Displacement and Sequence Check Metrics

Table 2

Performance analysis of the FSC2 keyframe extraction algorithm and existing algorithms in terms of the reduction rate and accuracy on Pheonix4 and How2Sign datasets.

Dataset	S	n	Method	m	R (%)	A (%)
Pheonix4 (train)	5670	798629	Pixel difference [13]	203650	74.5	67.3
			Gradient based [27]	214032	73.2	66.4
			Zernik’s moment [35]	492754	38.3	70.4
			Sampling [29]	238790	70.1	60.3
			FSC2 (proposed)	206823	74.1	83.1
How2Sign (test)	32	101793	Pixel difference [13]	30233	70.3	69.7
			Gradient based [27]	35322	65.3	70.1
			Zernike’s moment [35]	60974	40.1	74.4
			Sampling [29]	28400	72.1	67.3
			FSC2 (proposed)	24995	75.4	84.2

S denotes the total input video, n denotes the total frame count before reduction, m denotes the keyframe count after reduction, R is the reduction rate, and A is the average accuracy. The highlighted values shows that the FSC2 algorithm gives a higher reduction rate and accuracy.