Learning Model Depth Map Correction for Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recognition processing on image data from sensors can be compromised by optical and electrical factors, leading to deteriorated image quality and potential leakage of sensitive information such as faces or fingerprints.
Innovation Solution
An information processing system that includes a specifying unit to identify correction target pixels in depth maps or image data using a learning model and a correction unit to preprocess these pixels before recognition processing, improving image quality and securing sensitive information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition image data is used for recognition processing, then recognition accuracy is improved, but sensitive information such as faces and fingerprints may leak to outsiders
Solution Approach 1:
The patent extracts only the necessary recognition features from the image data while removing or obscuring sensitive information. The learning model is trained to identify target objects based on non-sensitive characteristics, separating useful recognition information from privacy-sensitive data.
Solution Approach 2:
The patent applies different processing qualities to different regions of the image data. Sensitive regions (faces, fingerprints) are processed with lower quality or obscured, while non-sensitive regions maintain higher quality for recognition purposes. This creates spatially varying information quality throughout the image.
2Speed
If image data is directly used for recognition processing, then processing speed is maintained, but image quality deteriorates due to optical and electrical factors
Solution Approach 1:
The patent performs preliminary correction processing on the image data before recognition processing. A learning model pre-processes the image to correct optical and electrical degradation, improving image quality in advance. This preliminary action reduces the need for repeated processing and maintains efficient overall processing speed.
Solution Approach 2:
The patent introduces a learning model as an intermediary between the raw image data and the recognition processing. This intermediary component corrects image quality issues caused by optical and electrical factors, acting as a mediator that improves image quality without significantly increasing overall processing time.
3Measurement precision
If preprocessing is applied to image data before recognition processing, then image quality is improved, but processing complexity increases
Solution Approach 1:
The patent employs a self-service learning model that automatically learns and adapts to the specific characteristics of the imaging system and environmental conditions. The model self-adjusts correction parameters based on training data, eliminating the need for manual calibration and reducing operational complexity despite the added processing step.
Data Source
AI summary
Before recognition processing is performed, preprocessing is performed on image data acquired by a sensor or image data obtained by converting the image data. An information processing system according to an embodiment includes a specifying unit (201) that specifies a correction target pixel in a depth map using a first learning model and a correction unit (202) that corrects the correction target pixel specified by the specifying unit.