Scene Change Detection via Coding Cost Differences in Video Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting scene changes in streamed video are inadequate, particularly in identifying scene changes that occur within normal Group of Pictures (GOP) lengths, leading to false positives and missed detections, which affects video quality assessment and error resilience in error-prone networks.
Innovation Solution
A method that calculates the difference in coding costs of macro-blocks across successive pictures, identifies new scenes based on threshold criteria, and applies motion compensation to account for misalignment and varying macro-block types, allowing for accurate scene change detection without decoding the video stream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scene change detection is performed using traditional methods in streamed video, then detection can be implemented, but false positives and missed detections occur particularly within normal GOP lengths, reducing measurement precision
Solution Approach 1:
The patent changes the parameter being measured from simple pixel difference to coding cost difference. By calculating the difference in coding costs between macro-blocks at the same position in successive pictures, the method achieves more reliable scene change detection with fewer false positives and missed detections, particularly within normal GOP lengths.
2Measurement precision
If full-reference perceptual models are used for video quality assessment, then quality evaluation accuracy is improved, but computational complexity increases making them unsuitable for large-scale network monitoring
Solution Approach 1:
The patent extracts only the essential information needed for scene change detection from the video stream - specifically the coding cost data from the bit stream. By taking out only this critical parameter rather than performing full video decoding and analysis, the method achieves accurate scene change detection with significantly reduced computational complexity, making it suitable for large-scale network monitoring.
3Reliability
If I or IDR pictures are inserted periodically to refresh video, then error resilience is improved, but encoding cost increases significantly
Solution Approach 1:
The patent performs preliminary scene change detection using coding cost differences before encoding decisions are made. By identifying scene changes in advance through this low-cost analysis, the system can strategically insert I or IDR pictures only when necessary for error resilience, avoiding unnecessary encoding of full I-pictures and thus reducing overall encoding cost while maintaining reliability.
Data Source
AI summary
There are provided a plurality of methods for detecting a scene change in a streamed video, the streamed video comprising a series of pictures. An example method comprises calculating, for a plurality of positions, a difference between the costs of coding macro-blocks at the same position in successive pictures. The method further comprises identifying a new scene when the sum of the differences for a plurality of positions meets a threshold criterion. There is further provided a method of determining the perceptual impact of a packet loss on a streamed video the method comprising: identifying a packet loss; and determining if the lost packet contained information relating to a picture at the start of a new scene, wherein a new scene is detected using one of the methods disclosed herein.