View Synthesis Depth Map Reliability Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current image synthesis algorithms for immersive videos, such as 360° and 180° videos, face significant quality issues due to errors in depth maps, which are introduced during capture, calculation, or compression, leading to blurry or gross defects in synthesized views, affecting viewer immersion.

Innovation Solution

A method that conditionally modifies projected depth map pixels based on reliability information, using weighting and confidence levels to correct errors, ensuring that only necessary modifications are made to reduce artifacts in the synthesized image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If depth maps are projected to synthesize intermediate views, then view synthesis capability is achieved, but projection errors and depth map defects cause significant quality degradation in synthesized images

Engineering Contradiction:
Improveview synthesis capabilityVSAvoidsynthesized image quality
Core Design Contradiction:
Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent applies preliminary filtering and confidence assessment to depth maps before they are used in view synthesis. By pre-identifying and correcting unreliable depth values based on confidence metrics, the system prepares the depth data in advance to minimize projection errors and quality degradation in the final synthesized views.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces confidence metrics and reliability assessment mechanisms as intermediaries between the depth maps and the view synthesis process. These confidence values act as mediators to weight and select depth information, filtering out erroneous depth values before they contribute to synthesis artifacts, thereby improving overall image quality.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If depth map errors are corrected through filtering, then synthesized image quality improves, but processing complexity and computational cost increase

Engineering Contradiction:
Improvesynthesized image qualityVSAvoidprocessing complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality assessment by computing confidence metrics for individual depth values rather than uniformly processing entire depth maps. This localized approach allows the system to selectively refine only the unreliable depth values while leaving confident depth information unchanged, reducing unnecessary processing complexity while maintaining image quality.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the parameter representation by introducing confidence metrics alongside depth values. This additional parameter enables intelligent differentiation between reliable and unreliable depth information, allowing selective processing that improves quality without proportionally increasing overall processing complexity.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3878170B1View synthesis
Publication Date: 2024.08.14 ORANGE SA
  • EP3878170B1 patent drawingFigure 1
  • EP3878170B1 patent drawingFigure 2
  • EP3878170B1 patent drawingFigure 3A~3C

AI summary

The invention relates to a method for synthesizing an image (lsth) of a view from images of N (N>2) views (T1, D1; T2, D2;...; TN, DN), implemented by an image synthesis device comprising the following: - projecting (S1), to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively; - for at least one given pixel of at least one projected depth map (Dj v), for which a depth value has been associated on completion of projection, modifying (S2) the depth value of said at least one given pixel if an item of reliability information (Fij) associated with said depth value is at a certain value (V2), said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map (Dj v).