Inference Target Region Selection for Image Processing Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing systems face challenges in maintaining image processing accuracy when resizing image data for input into machine learning models, leading to decreased noise reduction accuracy and increased initialization time for inference.

Innovation Solution

An information processing apparatus that sets up an inference target region in an input image and selects a suitable model for inference based on the size of this region, from a plurality of models with different input data sizes, to minimize initialization time and maintain processing accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the size of image data is changed to match model input requirements, then the image data can be processed by the model, but the feature of the image changes and processing accuracy decreases

Engineering Contradiction:

Improvemodel compatibilityVSAvoidimage processing accuracy

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The image is divided into multiple regions of interest (ROIs) with different sizes. Each ROI is processed by a model that matches its size, avoiding the need to resize the entire image and preserve local features while achieving model compatibility

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different regions of the image are processed with different model sizes according to their specific requirements. Important regions use larger models for higher accuracy, while less critical regions use smaller models, optimizing both accuracy and efficiency

Inventive Principle:

Principle #3Local quality

2Reliability

If a model is initialized to execute inference on target image data, then the model can process the image, but initialization time increases and waiting time becomes apparent

Engineering Contradiction:

Improveinference execution capabilityVSAvoidinitialization time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

Multiple models with different sizes are pre-initialized and kept ready before inference is needed. When an ROI is to be processed, a suitable pre-initialized model is selected and applied immediately, eliminating initialization waiting time

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Multiple models with different input size capabilities are maintained in the system. This multi-functionality allows the system to handle various ROI sizes without reinitialization, as appropriate models are already available

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20250077912A1Information processing apparatus, method, and storage medium

Publication Date: 2025.03.06 CANON KK

AI summary

An information processing apparatus includes at least one memory storing instructions, and at least one processor that, upon execution of the stored instructions cause the at least one processor to set up, in an input image, an inference target region subjected to an inference by a model established based on machine learning and decide, according to a size of the set inference target region, a model to be applied to an inference in which the input image is set as input data from among a plurality of models which have mutually different sizes of input data and on which an initialization for initiating a state in which the inference is executable on the input data is implemented.