On-Device Image Grouping via Local ML Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for a reliable and efficient method to automate the identification and grouping of images on end-user devices, such as smartphones and tablets, without relying on cloud-based services, which may be less convenient, less secure, and less responsive, especially due to privacy concerns and limited network bandwidth.
Innovation Solution
A system is described that includes a processor, memory, and a storage medium with an image manager, image group classifier, and ranking engine to classify images locally using machine learning-based grouping classifiers, specifically trained facial image classifiers, to determine embeddings, calculate confidence scores, and rank images for display on a graphical user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If cloud-based image classification services are used, then image processing capability is improved, but user privacy and security deteriorate
Solution Approach 1:
The patent introduces an on-device image classification system that acts as an intermediary between the user's photo library and cloud services. The classifier runs locally on the end-user device, processing images without requiring cloud upload, thus maintaining privacy while enabling intelligent organization and search capabilities.
2Power
If cloud-based image classification is used, then processing power is improved, but network bandwidth consumption increases
Solution Approach 1:
The system implements self-service by enabling the end-user device to perform image classification independently using on-device machine learning models. The device processes and classifies images locally without requiring network communication, eliminating bandwidth consumption while maintaining processing capabilities through efficient neural network implementations.
3Measurement precision
If cloud-based image classification is used, then classification accuracy is improved, but response time deteriorates
Solution Approach 1:
The system applies preliminary action by pre-training and deploying optimized machine learning classification models directly on the end-user device before actual image classification is needed. This allows immediate local inference without cloud latency, achieving both high accuracy through pre-trained models and fast response times through on-device execution.
Data Source
AI summary
Systems, methods, and data storage devices for image grouping in an end user device using trained machine learning group classifiers are described. The end user device may include an image group classifier configured to classify new image data objects using an image classification algorithm and set of machine learning parameters previously trained for a specific image group. The end user device may determine embeddings that quantify features of the target image object and use those embeddings and the image group classifier to selectively associate group identifiers with each new image data object received or generated by the end user device. Calibration, including selection and training, of the image group classifiers and ranking of classified images are also described.