CNN-LSTM Learning Approach-Based Complexity Reduction for High-Efficiency Video Coding Standard

Table 2

Performances evaluation [CNN-LSTM versus deep CNN].

Class	Sequence	Deep CNN [9]			Proposed CNN-LSTM
		BD rate (%)	BD-PSNR (dB)	ΔT (%)	BD rate (%)	BD-PSNR (dB)	ΔT (%)
A (2560 × 1600)	PeopleOnStreet	1.20	−0.051	−50.67	1.70	−0.017	−48.88
	Traffic	1.49	−0.041	−57.90	1.53	−0.059	−66.38
B (1920 × 1080)	Kimono	1.38	−0.044	−43.26	1.65	−0.052	−47.77
	ParkScene	1.43	−0.041	−64.14	2.79	−0.081	−70.82
	Cactus	2.44	−0.047	−52.57	1.73	−0.033	−53.85
	BQTerrace	2.22	−0.034	−58.43	1.75	−0.030	−65.62
	BasketballDrive	2.28	−0.051	−51.30	2.02	−0.045	−52.77
C (832 × 480)	BasketballDrill	1.43	−0.052	−53.54	1.67	−0.061	−48.23
	BQMall	2.24	−0.085	−52.25	1.38	−0.090	−48.07
	PartyScene	1.48	−0.057	−51.54	0.96	−0.038	−59.30
	RaceHorses	1.41	−0.053	−42.22	1.47	−0.055	−54.32
D (416 × 240)	BasketballPass	1.85	−0.083	−52.42	1.26	−0.056	−56.67
	BQSquare	2.09	−0.073	−52.79	1.27	−0.046	−60.79
	BlowingBubbles	1.71	−0.061	−46.55	0.97	−0.034	−50.14
	RaceHorses	1.32	−0.058	−38.01	1.60	−0.018	−50.33
E (1280 × 720)	FourPeople	1.06	−0.029	−67.54	2.71	−0.071	−72.42
	Johnny	3.99	−0.083	−69.66	2.46	−0.083	−73.59
	KristenAndSara	1.31	−0.082	−67.20	2.93	−0.094	−74.91
Average		1.80	−0.057	−53.99	1.78	−0.053	−58.60