Pixel Consistency Data for Computer Vision Reliability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Image processing for computer vision systems can have unpredictable effects, leading to unreliable decision processes, particularly in identifying target objects and navigating autonomous vehicles.

Innovation Solution

A method that involves receiving image data, applying image processing to produce a processed image and consistency data, which indicates the consistency of the processed image with the original data, and using both as inputs for a computer vision decision process to improve reliability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If image processing is applied to improve image data quality, then image quality is improved, but the reliability of computer vision decision processes becomes unpredictable

Engineering Contradiction:
Improveimage qualityVSAvoiddecision process reliability
Core Design Contradiction:
Manufacturing precisionVSReliability

Solution Approach 1:

The processed image is divided into multiple regions, each with its own consistency metric indicating the reliability of image processing in that region. The computer vision system then selectively uses image data from regions with high consistency while relying more on raw image data from regions with low consistency, thereby maintaining decision reliability while preserving image quality improvements where valid.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A consistency metric layer is introduced as an intermediary between the processed image and the computer vision decision process. This consistency data acts as a mediator that guides the computer vision system on which processed image regions to trust, allowing the system to differentiate between reliable and unreliable processed regions without discarding the benefits of image processing entirely.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If image processing is applied to correct defects and inconsistencies, then image data accuracy is improved, but the complexity of the processing system increases

Engineering Contradiction:
Improveimage data accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system pre-calculates and stores consistency metrics for each region of the processed image before the computer vision decision process occurs. This preliminary action allows the computer vision system to quickly query reliability information without performing complex real-time analysis, thereby improving image data accuracy while minimizing the complexity added to the decision process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Rather than applying uniform complex processing to the entire image, the system applies different levels of processing and consistency checking to different regions based on their specific characteristics. Regions with high processing reliability receive full processing benefits, while regions with low reliability rely more on raw data, thereby improving overall accuracy without unnecessarily complicating the entire system.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS10325359B2Pixel consistency
Publication Date: 2019.06.18 ARM LTD
  • US10325359B2 patent drawing
  • US10325359B2 patent drawing
  • US10325359B2 patent drawing

AI summary

According to an aspect of the present disclosure, there is provided a method of image processing. The method comprises receiving image data. Image processing is applied to the image data, whereby to produce a processed image. The method then comprises producing consistency data associated with the image processing, wherein the consistency data is indicative of a consistency of at least one region of the processed image with the received image data. Both the processed image and the consistency data are used as inputs to a decision process of a computer vision system.