Deep Neural Network for Adjusting Pictorial Depth Cues

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for generating images with enhanced depth perception using depth maps suffer from low accuracy, leading to incorrect depth information being provided to viewers.

Innovation Solution

An electronic device and method utilizing a deep neural network comprising multiple neural networks and controllable conversion modules to adjust pictorial depth cues such as blur, contrast, and sharpness based on control parameters, without relying on depth maps, to enhance depth perception.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If depth maps are used to generate images with enhanced depth perception, then depth enhancement can be achieved, but accuracy deteriorates leading to incorrect depth information

Engineering Contradiction:
Improvedepth perception accuracyVSAvoiddepth map accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent extracts and directly processes pictorial depth cues (blur, contrast, sharpness) from the input image without relying on intermediate depth maps. By taking out the essential depth-related visual features and processing them directly through neural networks, the method eliminates the accuracy limitations of depth map generation while preserving depth enhancement capability

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces controllable conversion modules as intermediaries between the input image and output image. These modules convert feature maps from hidden layers of neural networks and apply controlled adjustments to pictorial depth cues, enabling precise regulation of depth enhancement effects without the inaccuracies associated with depth map-based approaches

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If deep neural networks with multiple neural networks and controllable conversion modules are used, then depth perception accuracy is improved, but device complexity increases

Engineering Contradiction:
Improvedepth perception accuracyVSAvoidneural network structure complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the deep neural network into three specialized neural networks (first, second, and third) with distinct functions, plus controllable conversion modules. The first neural network extracts features, the second processes spatial relationships, and the third generates output. This segmentation allows each component to be optimized for its specific task, improving overall accuracy while making the complex system more manageable and interpretable

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The controllable conversion modules serve multiple functions: they convert feature maps from different neural networks, apply controlled transformations to pictorial depth cues, and integrate results from multiple processing paths. This multi-functionality reduces the need for separate dedicated components for each operation, thereby managing device complexity while maintaining high depth perception accuracy

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

Data Source

PatentEP4502945B1Electronic device and method for generating image in which depth recognized by viewer is reinforced
Publication Date: 2026.04.22 SAMSUNG ELECTRONICS CO LTD
  • EP4502945B1 patent drawingFigure 1
  • EP4502945B1 patent drawingFigure 2
  • EP4502945B1 patent drawingFigure 3

AI summary

An electronic device for generating an image with an enhanced depth as perceived by a viewer includes at least one memory storing instructions, and at least one processor configured to execute the instructions to receive an input image, receive a control parameter for adjusting a first level of at least one pictorial depth cue included in the input image, and generate an output image including the at least one pictorial depth cue at a second level that is adjusted from the first level based on the control parameter using a deep neural network.