Imaging Control for Privacy-Preserving Lensless Training Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training images captured by lensless cameras, such as multi-pinhole cameras, are difficult for humans to recognize, leading to inaccurate correct answer information and degradation of machine learning model performance, making it challenging to improve recognition accuracy while protecting privacy.
Innovation Solution
A system that displays first correct answer information on a display device, captures a blurred image using an imaging device, and generates second correct answer information based on the captured image, creating a data set including the training image and correct answer information for machine learning model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a lensless camera captures training images to protect privacy, then privacy protection is improved, but recognition accuracy of the machine learning model deteriorates due to inaccurate correct answer information
Solution Approach 1:
The system segments the training process into two distinct parts: (1) capturing blurred training images with the lensless camera to protect privacy, and (2) capturing sharp reference images with a conventional camera to generate accurate correct answer information. This segmentation allows each camera type to perform its optimal function without compromise.
Solution Approach 2:
The conventional camera acts as an intermediary device that captures sharp reference images of the same display content. These reference images serve as mediators to generate accurate correct answer information, which is then paired with the blurred training images from the lensless camera, enabling accurate model training without exposing private information.
2Object-affected harmful factors
If blurred images are used for training to protect privacy, then privacy protection is improved, but correct answer information accuracy deteriorates
Solution Approach 1:
The system separates the image capture function into two modes: blurred image capture for privacy protection and sharp reference image capture for accurate annotation. This segmentation ensures that privacy is maintained in the training data while accurate correct answer information is obtained through the reference imaging device.
Solution Approach 2:
The system creates a copy of the display content using a conventional camera to generate sharp reference images. These copies serve as the basis for creating accurate correct answer information, eliminating the need to derive annotations directly from blurred images while maintaining privacy protection.
3Measurement precision
If accurate correct answer information is generated from blurred images, then recognition accuracy is improved, but privacy protection deteriorates
Solution Approach 1:
The system divides the imaging function into two specialized components: a lensless camera that captures only blurred images for privacy protection, and a conventional camera that captures sharp reference images for accurate annotation. This segmentation allows accurate correct answer information to be generated without compromising the privacy-protecting blurred training images.
Solution Approach 2:
The conventional camera serves as an intermediary that captures sharp reference images to generate accurate correct answer information. This intermediary approach allows the system to obtain precise annotation data without directly processing or exposing the content of blurred training images, thereby maintaining privacy protection while improving recognition accuracy.
Data Source
AI summary
An imaging control device includes a correct answer information acquiring part that acquires first correct answer information corresponding to a training image of a machine learning model from a first storage part, a correct answer image display control part that causes a display device to display a first correct answer image based on the first correct answer information, an imaging control part that causes an imaging device that acquires a blurred image to capture the first correct answer image displayed on the display device and acquires a second correct answer image, a correct answer information generating part that generates second correct answer information based on the second correct answer image, and a storage control part that stores a data set including a set of the training image and the second correct answer information in a second storage part.