Facial Geometry Encoding for Compatible Low-Bandwidth Video Re-Enactment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing face re-enactment techniques lack compatibility across different neural network implementations due to unique configurations, leading to difficulties in interoperation and inefficient data transfer.

Innovation Solution

Encoding geometric information, such as feature point locations, instead of entire images, to facilitate universal reuse and reduce data volume, while ensuring natural output video generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If entire images are encoded for face re-enactment, then image quality is preserved, but data transfer volume increases

Engineering Contradiction:
Improveimage qualityVSAvoiddata transfer volume
Core Design Contradiction:
Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential geometric information (feature point locations, facial landmarks, pose parameters) from complete images. Instead of encoding entire image data, the system identifies and encodes only the critical geometric attributes that define facial geometry, thereby reducing data volume while preserving the necessary information for face re-enactment.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the face into discrete feature points and geometric elements. By dividing the continuous image data into distinct geometric components (landmarks, contours, pose parameters), the system enables selective encoding of only the essential geometric information needed for reconstruction, rather than transmitting complete image data.

Inventive Principle:
Principle #1Segmentation

2Reliability

If different neural network configurations are used, then model specialization improves, but compatibility across implementations deteriorates

Engineering Contradiction:
Improvemodel specializationVSAvoidcompatibility
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent establishes a universal geometric information encoding format that can be used across different neural network implementations. By defining a standardized structure for representing facial geometry (feature points, landmarks, pose parameters), the system enables compatibility between different models while allowing each implementation to specialize in its own processing approach.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent changes the representation parameters from raw image pixels to standardized geometric attributes. This parameter transformation creates a model-agnostic intermediate representation that can be processed by various neural networks while maintaining consistency in how facial geometry is described and transmitted.

Inventive Principle:
Principle #35Parameter changes

3Quantity of substance

If geometric information is encoded instead of entire images, then data transfer volume reduces, but information completeness may be compromised

Engineering Contradiction:
Improvedata transfer volumeVSAvoidinformation completeness
Core Design Contradiction:
Quantity of substanceVSLoss of information

Solution Approach 1:

The patent applies partial action by encoding only the essential geometric information needed for face re-enactment rather than complete image data. By identifying and encoding only the critical feature points and geometric attributes that define facial geometry, the system achieves sufficient information completeness for the specific application while significantly reducing data volume.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4654579A1Encoding device, decoding device, encoding method, and decoding method
Publication Date: 2025.11.26 PANASONIC INTELLECTUAL PROPERTY CORP OF AMERICA
  • EP4654579A1 patent drawingFigure 1
  • EP4654579A1 patent drawingFigure 2
  • EP4654579A1 patent drawingFigure 3

AI summary

An encoder (100) includes memory (152) and circuitry (151) coupled to the memory (152). Using the memory (152), the circuitry (151): encodes at least one fundamental image for use in displaying a video; and encodes, as information corresponding to each of images of the video, geometric information indicating geometric attributes within a region including a face of a person.