Compressed-Domain Camera Control Without Image Reconstruction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Compressive sensing capturing devices face challenges in implementing automatic camera features like autofocus due to the computationally intensive reconstruction required to retrieve the original image from compressed measurements, which is slow and battery inefficient.

Innovation Solution

Employing a machine learning algorithm based on an artificial neural network, specifically a convolutional neural network, to directly measure image features from compressed sensing data, allowing for faster and more efficient automatic adjustments of device attributes such as autofocus without reconstructing the original image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If compressive sensing is used to reduce measurements and storage, then battery usage and storage requirements are reduced, but computationally intensive reconstruction is required which is slow and battery inefficient

Engineering Contradiction:
Improvebattery usageVSAvoidreconstruction speed
Core Design Contradiction:
Use of energy by moving objectVSProductivity

Solution Approach 1:

The patent extracts and processes only the essential features needed for automatic camera controls from the compressed measurements, rather than performing full image reconstruction. This selective extraction of relevant information (edge detection, gradient calculation, focus metric computation) from the compressed domain eliminates the computationally intensive reconstruction step while maintaining the ability to perform autofocus and other automatic functions.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent inverts the traditional workflow by performing image processing operations directly in the compressed measurement domain rather than reconstructing the image first. Instead of following the conventional path of compressed sensing → full reconstruction → image processing, the patent implements image processing → compressed sensing operations, allowing automatic camera controls to be executed on the compressed data itself.

Inventive Principle:
Principle #13The other way round (Inversion)

2Quantity of substance

If compressive sensing is used to capture images with fewer measurements, then storage requirements are reduced, but the original image is not available for direct processing

Engineering Contradiction:
Improvestorage requirementsVSAvoidimage processing capability
Core Design Contradiction:
Quantity of substanceVSEase of operation

Solution Approach 1:

The patent introduces an intermediary processing stage that operates on compressed measurements to extract image attributes and control parameters. This intermediary layer includes operations such as computing gradients, detecting edges, and calculating focus metrics directly from the compressed data, serving as a bridge between the compressed sensing domain and the automatic camera control functions without requiring full image reconstruction.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If traditional compressive sensing reconstruction is performed to retrieve the original image, then complete image data is obtained, but the process is computationally intensive and slow

Engineering Contradiction:
Improveimage data completenessVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies partial action by performing only the necessary image processing operations needed for automatic camera controls on the compressed measurements, rather than completing the full reconstruction process. This partial processing approach extracts sufficient information for autofocus, exposure, and other automatic functions without the excessive computational burden of obtaining the complete reconstructed image.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3349359B1Compressive sensing capturing device and method
Publication Date: 2024.10.02 SONY GROUP CORP
  • EP3349359B1 patent drawingFigure 1~2
  • EP3349359B1 patent drawingFigure 3
  • EP3349359B1 patent drawingFigure 4~5

AI summary

A compressive sensing capturing device has circuitry, which is configured to obtain compressive sensing image data; and to set a device attribute based on image attribute data, wherein the image attribute data are based on a machine learning algorithm performed in the compressing domain on the obtained compressive sensing image data.