Neural Network Visual Artifact Detector for Video Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Service providers face challenges in detecting video artifacts, such as macroblocking, in video streams due to data loss and encoding/decoding errors, which are often not visible to users and require manual scanning or conventional reference-based detection methods.
Innovation Solution
An automated system using a neural network to analyze decoded video frames, detecting edges and characterizing potential macroblocks, trained with user ratings to determine the likelihood of macroblocking artifacts without needing a reference video.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual scanning by personnel is used to detect video artifacts, then detection accuracy is improved, but labor cost and time consumption increase significantly
Solution Approach 1:
The patent replaces manual mechanical scanning by personnel with an automated electronic detection system that uses digital signal processing and machine learning algorithms to identify video artifacts, thereby eliminating labor-intensive operations while maintaining high detection accuracy
Solution Approach 2:
The detection system performs self-monitoring of video streams through automated artifact detection algorithms that continuously analyze video data without requiring human intervention, enabling the system to service itself in detecting quality issues
2Reliability
If automated detectors are deployed at various network locations, then detection coverage is improved, but system complexity and infrastructure requirements increase
Solution Approach 1:
The patent creates a universal artifact detection system that can be deployed at any network location with consistent functionality, using standardized processing algorithms that work across different video streams and network positions without requiring location-specific customization
Solution Approach 2:
The system uses intermediate processing layers that sit between video encoding and delivery, acting as mediators that detect artifacts without requiring complex infrastructure changes at network locations, simplifying deployment while maintaining comprehensive coverage
3Measurement precision
If conventional reference-based detection methods are used, then artifact detection capability is improved, but adaptability to resolution changes and decoding protocols decreases
Solution Approach 1:
The patent implements dynamic detection algorithms that automatically adjust to different video resolutions and decoding protocols by learning from training data representing various video conditions, enabling the system to adapt its detection parameters based on the specific video stream being analyzed
Solution Approach 2:
The system changes its detection parameters dynamically based on the input video characteristics, adjusting sensitivity thresholds and analysis methods according to resolution, codec type, and stream conditions, thereby maintaining high detection capability across diverse video formats
Data Source
AI summary
A method is presented for detecting visual artifacts such as macroblocking artifacts appearing in a rendering of a video stream. A decoded video stream that includes a plurality of frames is received. Edges appearing in the plurality of frames are detected and data characterizing the edges is directed to a neural network. The neural network determines a likelihood that a macroblocking artifact occurs in the video stream using the neural network. The neural network is trained using training data that includes (i) one or more video sequences having known macroblocking artifacts whose size and duration are known and (ii) user ratings indicating a likelihood that users visually perceived macroblocking in each of the one or more video sequences when each of the one or more video sequences are rendered.