Image Recognition Program Comparison for Selective Image Saving

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image recognition programs, particularly those using deep neural networks, face challenges in performance degradation during updates, leading to frequent differences in detection results, making it difficult to determine which images to save for evaluation, resulting in a large number of unnecessary targets.

Innovation Solution

An image recognition device and method that includes a storage unit for current and new programs, a difference extraction unit to identify discrepancies, and a transmission determination unit to determine which images to save based on the status of these differences, using time-series and spatial filters to prioritize important discrepancies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If all images with detection differences are saved for evaluation, then the evaluation comprehensiveness is improved, but the storage and communication resources are excessively consumed

Engineering Contradiction:
Improveevaluation comprehensivenessVSAvoidnumber of images to save
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential difference information between current and new program detection results, rather than saving all images with any difference. The difference extraction unit identifies specific detection result differences, and the transmission determination unit uses this extracted difference information to selectively determine which images require saving, thereby reducing the quantity of saved images while maintaining evaluation effectiveness.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different evaluation criteria to different types of detection differences. The transmission determination unit evaluates the importance of each detected difference based on its characteristics, and only selects images with significant differences for saving. This local quality approach ensures that evaluation resources are concentrated on the most critical cases rather than treating all differences equally.

Inventive Principle:
Principle #3Local quality

2Reliability

If verification is performed using a large number of test patterns in virtual environment and real machine verification, then the program safety is improved, but the time and human resources required are excessively increased

Engineering Contradiction:
Improveprogram safetyVSAvoidverification time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary difference extraction and evaluation before actual image saving and detailed verification. The difference extraction unit pre-identifies detection result differences, and the transmission determination unit pre-evaluates their importance, so that only images requiring verification are selected for saving and subsequent real-machine verification, significantly reducing the time and resources needed for the verification process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary automated determination mechanism between virtual environment testing and real-machine verification. The transmission determination unit acts as an intermediary that automatically filters and selects critical test cases based on extracted difference information, reducing the need for extensive manual verification while maintaining program safety.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If deep neural network programs are updated to improve recognition performance, then the recognition accuracy is improved, but the detection result consistency deteriorates

Engineering Contradiction:
Improverecognition accuracyVSAvoiddetection result consistency
Core Design Contradiction:
Measurement precisionVSStability of the object's composition

Solution Approach 1:

The patent implements a feedback mechanism that compares detection results between current and new programs. The difference extraction unit continuously monitors detection result changes after program updates, and the transmission determination unit uses this feedback to identify and save images showing significant performance degradation or inconsistent detection behavior, enabling targeted verification and model refinement.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4645254A1Image recognition device and saved image determination method
Publication Date: 2025.11.05 ASTEMO LTD
  • EP4645254A1 patent drawingFigure 1
  • EP4645254A1 patent drawingFigure 2
  • EP4645254A1 patent drawingFigure 3

AI summary

This image recognition device, which performs object detection by using an image recognition program with respect to an input image, comprises: a storage unit that stores a current program, which is a conventional image recognition program, and a new program, which is a new image recognition program; a difference extraction unit that extracts a detection difference which is a difference between an object detection result of the current program and an image recognition result of the new program with respect to the same input image; a transmission determination unit that determines a saving need as to whether or not to save the input image on the basis of a status of appearance of the detection difference; and a saving unit that outputs the input image that the transmission determination unit has determined to save to the outside of the image recognition device, or that saves the input image that the transmission determination unit has determined to save in the image recognition device.