Inference Model Adaptation for Image Restoration Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image restoration techniques using AI often incorrectly detect normal areas as abnormal due to low restoration extent, especially in environments different from the training environment, leading to suboptimal performance in abnormality detection.
Innovation Solution
An apparatus and method that identifies areas in images where the inference performance by a learning model is below a threshold, collects similar images, and uses these images to improve inference results in a test environment by updating the model parameters based on additional learning, utilizing a convolutional neural network (CNN) and a cloud-server edge-device system for image degradation restoration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image restoration is performed using a learning model trained in a training environment, then the model can restore images in the training environment, but the restoration performance degrades when applied to test environments with different conditions
Solution Approach 1:
The patent applies preliminary action by pre-collecting multiple images with different degradation types and conditions to form an extended training dataset before model training. This allows the model to be exposed to diverse environmental conditions in advance, improving its adaptability to test environments without requiring retraining when environmental conditions change.
Solution Approach 2:
The patent implements parameter changes by systematically varying degradation parameters (noise levels, blur kernels, compression ratios) and environmental conditions during data preparation and model training. This enables the model to learn robust restoration capabilities across different parameter settings, resolving the contradiction between maintaining high restoration performance and adapting to varying environmental conditions.
2Measurement precision
If additional learning is performed using images from the test environment, then the performance of abnormality detection is enhanced, but the complexity of the system increases
Solution Approach 1:
The patent applies partial action by selectively performing additional learning only on images identified as having low restoration extent or high abnormality suspicion. Rather than processing all images through the complete additional learning pipeline, the system applies the enhanced learning process only where needed, improving abnormality detection accuracy while controlling system complexity through conditional execution.
3Reliability
If the restoration extent of an image is low, then normal areas may be erroneously detected as abnormal, but increasing restoration extent may miss actual abnormal areas
Solution Approach 1:
The patent implements feedback by using the restoration result quality as a control signal to determine whether additional learning should be applied. When restoration extent is low, the system feeds back this information and triggers additional learning on that specific image or region. This closed-loop approach ensures that images with low restoration quality receive targeted improvement, reducing false abnormal detections while preserving true abnormal area detection.
Data Source
AI summary
An apparatus includes at least one processor, and a memory coupled to the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to identify a partial image corresponding to an area in an image in which performance of inference by a learning model that performs predetermined inference on an input image is less than or equal to a threshold, collect a similar image similar to the identified partial image, and based on additional images including the collected similar image, improve a result of the inference by the learning model targeted at a test environment different from a training environment of the learning model.