Model-Based Video Correction for Packet Loss Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression techniques, such as frame duplication and blurring, are inadequate in maintaining video quality over low bit rate connections, particularly in IPTV and iTV networks, where unrecoverable packet loss leads to noticeable defects and a degraded consumer experience.
Innovation Solution
Implementing model-based video correction methods that recognize video compression artifacts using stored models to synthesize and correct defects in video streams, particularly for objects like faces and backgrounds, enhancing image quality without attempting to replicate the original source.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional concealment techniques (frame duplication, blurring, averaging) are used to mitigate packet loss defects, then the visible defects are reduced, but the video quality remains degraded and consumer experience is impaired
Solution Approach 1:
The patent creates synthetic copies of lost video frames by generating new pixel data based on learned patterns from uncompressed video sequences. Instead of merely duplicating existing frames or applying blur, the system synthesizes entirely new frame content that matches the expected visual patterns, thereby eliminating visible defects while maintaining high video quality.
Solution Approach 2:
The patent replaces conventional mechanical concealment methods (frame duplication, blurring, averaging) with a model-based synthesis approach. The system uses learned probability models of natural video patterns to generate synthetic frames, substituting simple signal processing operations with intelligent content generation that preserves visual quality.
2Reliability
If model-based video correction is implemented to synthesize high-quality images, then subjective video quality is significantly improved, but the system complexity increases
Solution Approach 1:
The patent performs preliminary learning of video patterns during an offline training phase using uncompressed video sequences. The system learns probability models of natural video patterns beforehand, so that during actual video transmission, the system can quickly apply these pre-learned models to synthesize corrected frames without complex real-time computation.
Solution Approach 2:
The patent changes the fundamental parameters of frame generation from copying existing frames to synthesizing new frames based on probability models. The system transforms video correction from a pixel-level manipulation problem to a statistical modeling problem, where learned patterns guide the generation of high-quality synthetic frames.
Data Source
AI summary
A media processor having a controller operable to recognize a portion of a video stream in an interactive television network having video compression artifacts corresponding to a stored model and perform model-based video correction of the portion recognized using synthetically generated images of objects in a captured video scene. Other embodiments are disclosed.