Image Recognition Processing with Adaptive Resolution Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Image recognition processing based on machine learning, such as deep learning, faces challenges with long processing times when dealing with high-resolution images, and reducing image resolution to improve speed results in insufficient recognition accuracy.

Innovation Solution

An image processing apparatus that performs recognition processing on both low and high-resolution images, determining the reliability of the results and adjusting the resolution accordingly to balance accuracy and speed, by executing recognition processing in a first image at a reduced resolution and a second image at a higher resolution based on predetermined standards.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If recognition processing is performed on high-resolution images, then recognition accuracy is improved, but processing time increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The image processing is segmented into multiple resolution levels. The system divides the recognition task into coarse-level processing (low resolution) for rapid classification and fine-level processing (high resolution) for detailed analysis, allowing parallel processing at different scales to balance speed and accuracy

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary recognition processing on low-resolution images before high-resolution processing. This preliminary action filters out obviously incorrect classifications and prepares candidate regions, so that subsequent high-resolution processing only needs to focus on specific areas of interest, reducing overall processing time

Inventive Principle:
Principle #10Preliminary action

2Productivity

If image resolution is decreased to improve processing speed, then processing time is reduced, but recognition accuracy becomes insufficient

Engineering Contradiction:
Improveprocessing speedVSAvoidrecognition accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts the processing resolution based on the complexity of the recognition task and the confidence level of preliminary results. For simple cases, low-resolution processing suffices; for complex or ambiguous cases, the system automatically transitions to high-resolution processing to maintain accuracy

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system introduces an intermediate processing stage that bridges low and high resolution processing. This intermediary layer performs feature extraction and candidate selection at medium resolution, acting as a bridge that prepares data for high-resolution analysis while filtering out unnecessary computations

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS20230245442A1Image processing apparatus, image processing method, and non-transitory computer readable medium
Publication Date: 2023.08.03 CANON KK
  • US20230245442A1 patent drawing
  • US20230245442A1 patent drawing
  • US20230245442A1 patent drawing

AI summary

An image processing apparatus configured to perform recognition processing for a target object in an input image, includes at least one memory storing a program, and at least one processor. The processor, by executing the program, causes the image processing apparatus to perform the recognition processing for the target object in a first image having a first resolution acquired based on the input image and acquire a first recognition result, determine whether or not a reliability of the first recognition result satisfies a predetermined standard, and perform the recognition processing for the target object in a second image which has a higher resolution than the first resolution acquired based on the input image, and acquire a second recognition result, in a case where it is determined that the reliability of the first recognition result does not satisfy the predetermined standard.