2004, 9/1 1 Optimal Content-Based Video Decomposition for Interactive Video Navigation Anastasios D. Doulamis, Member, IEEE and Nikolaos D. Doulamis, Member, IEEE 謝俊瑋
2004, 9/1 2 Introduction Demand of multimedia data transmitted over network is rapidly increasing. Using video file downloading or streaming are tedious and time consuming. Video summarization is not appropriate for interactive video navigation over network. Hierarchical content-based video decomposition for interactive video navigation.
2004, 9/1 3 Outline Efficiency of video content decomposition Video content decomposition Optimization method Tree node representation and transmission Detail estimation and regulation Experimental result Conclusion
2004, 9/1 4 Efficiency of video content decomposition Traditional video organization scheme: 2-level video content decomposition: --(1)
2004, 9/1 5 Efficiency of video content decomposition (2) Worst case of (1) is obtained when all shot classes have the same probability Worst case of 4-level decomposition is 1/16 of total number of frames
2004, 9/1 6 Video content decomposition Features: – Global frame properties (global-based descriptors): Color, texture and motion histogram – Object-based descriptors: Use segmentation algorithm: M-RSST Multiresolution implementation of recursive shortest spanning tree algorithm Average color components, region location, size
2004, 9/1 7 Video content decomposition (2) Extraction of shot/frame representatives
2004, 9/1 8 Video content decomposition (3) Extraction of shot/frame representatives (2)
2004, 9/1 9 Video content decomposition (4) Shot-class construction – Influence zone – Shot class – Shot representative
2004, 9/1 10 Video content decomposition (5) Frame-class construction – Respective frame representatives
2004, 9/1 11 Optimization method Guided random search procedure implemented by a genetic algorithm – Initial state – Fitness function
2004, 9/1 12 Optimization method (2) 1. Population performance evaluation 2. Next population construction phase – Select parents – Crossover – Mutation
2004, 9/1 13 Node representation and transmission Viewing elements of shot s Viewing element of show class Viewing element of frame class
2004, 9/1 14 Node representation and transmission (2)
2004, 9/1 15 Detail estimation and regulation Average transmitted information
2004, 9/1 16 Detail estimation and regulation (2) For the multiple paths, use the visual content complexity as probability and consider degree of interest
2004, 9/1 17 Detail estimation and regulation (3) If the number representative shots or frames, does not satisfy the user’s information needs? Maximum deviation class separation method
2004, 9/1 18 Experimental result
2004, 9/1 19 Conclusion The aforementioned difficulties are addressed using on a nonsequential (nonlinear) video navigation scheme. Improve of the GA