290 likes | 408 Views
Balanced Parallel Scheduling for Video Encoding with Adaptive GOP Structure. IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 24, NO. 12, DECEMBER 2013 Hsu- Feng Hsiao, Member, IEEE, and Chen-Tsang Wu. Outline. Introduction Related Work Methods Experimental Results.
E N D
Balanced Parallel Scheduling for Video Encoding with Adaptive GOP Structure IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 24, NO. 12, DECEMBER 2013 Hsu-Feng Hsiao, Member, IEEE, and Chen-Tsang Wu
Outline • Introduction • Related Work • Methods • Experimental Results
Introduction • Idea 1: • Scene change • Idea 2: • Scheduling
Related Work #1 • Scene change detection • Abrupt scene change • An abrupt scene change indicates that transition from one scene into another only spends a period of one frame. • Gradual scene change • A gradual scene change takes a period of several frames to complete a scene transition.
Related Work #2 • Pixel-based method • Sum of mean absolute difference (SMAD) • Block-based method • A scene change is declared if the ratio of similar blocks between two frames is greater than some threshold.
Related Work #3 • Frame-level parallelism
Method #1 • Sum of absolute temporal difference (SATD) • Where is the pixel value at location (x,y).
Method #2 • Sum of absolute spatial difference (SASD)
Method #3 • Sum of absolute spatial difference (SASD)
Method #5 • Abrupt Scene Change Detection Algorithm • If Ratio is greater than the predefined threshold Abrupt Scene Change
Method #6 • Ratio noise • Threshold = 1.4
Method #8 • 2nd derivative • Local maxima of the ratios shall be at the positions where the values are negative. • Threshold = -1.0
Method #9 • Gradual Scene Change Detection Algorithm • A gradual scene change takes a period of several frames • Ratio(n,m) is the ratio between frame n and frame m.
Method #10 • Beginning • Threshold = 0.7(0.3~1.0) • Ending • Convergence
Method #13 I : 120ms B : 290ms P : 360ms