Image Patch Privacy Protection With Neural Network Guidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional electronic devices may inadvertently disclose privacy objects in photos or images, leading to potential privacy infringements when shared on social networks or with other devices.
Innovation Solution
An electronic apparatus equipped with a neural network model processes images into patches, identifies privacy objects, and provides a graphical user interface to guide and apply protective measures such as blur, single-color, or emoji processing, and applies adversarial noise or watermark filters to prevent detection by other neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image processing is performed to protect privacy objects, then privacy protection effectiveness is improved, but image quality and detail are degraded
Solution Approach 1:
The patent applies different processing methods to different regions of the image based on object type. Privacy objects like faces and license plates receive strong protection (blur, pixelation, or replacement with emojis), while non-privacy regions maintain original quality. This localized approach ensures maximum privacy protection for sensitive areas while preserving overall image quality in non-sensitive areas.
Solution Approach 2:
The system dynamically adjusts protection parameters based on the type of privacy object detected. For example, faces may receive Gaussian blur with specific kernel sizes, while license plates may receive character replacement with emojis. The processing intensity, filter strength, and protection method are all parameter-adjusted according to the detected object category, balancing privacy protection with image quality preservation.
2Ease of operation
If multiple processing methods are provided for user selection, then ease of operation is improved, but device complexity increases
Solution Approach 1:
The system pre-configures multiple protection methods (blur, pixelation, emoji replacement, blacking out) and automatically detects privacy objects before presenting options to the user. This preliminary detection and preparation of processing methods allows the user to simply select from predefined options rather than configuring complex parameters, reducing operational complexity while maintaining flexibility.
Solution Approach 2:
The system automatically identifies privacy objects and applies appropriate protection methods based on user selection without requiring manual configuration. The automatic detection, classification, and application of protection methods reduces the burden on users, making the system easy to operate while the underlying complexity is handled automatically by the device.
3Measurement precision
If neural network models are used to identify privacy objects, then measurement precision is improved, but loss of information increases due to processing overhead
Solution Approach 1:
The image processing is divided into multiple stages: first, the image is segmented into regions of interest using the neural network to detect potential privacy objects; then, only those specific regions undergo detailed analysis and protection processing. This segmentation approach allows the neural network to focus computational resources on identifying privacy objects while minimizing processing of the entire image, thereby reducing overall processing time and information loss.
Solution Approach 2:
The system applies neural network processing selectively rather than uniformly across the entire image. It performs partial processing by focusing computational effort only on regions where privacy objects are likely to be present, based on preliminary detection. This partial action approach maintains high detection accuracy for privacy objects while significantly reducing the total processing time compared to full-image neural network analysis.
Data Source
AI summary
Disclosed is an electronic apparatus. The electronic apparatus may comprise: a memory storing an application and a neural network model; a display; and a processor which, by executing the application, differentiates an input image into a plurality of patches, obtains information about a privacy object included in the input image by inputting the plurality of patches to the neural network model, obtains an output image including a guide GUI for guiding the privacy object based on the obtained information, and controls the display to display the obtained output image.