On-Device Pet Image Clustering Using Face and Body Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pet clustering systems face challenges in achieving high precision and recall under diverse conditions, particularly with images of pet bodies, and they often rely on cloud-based infrastructure without considering privacy and connectivity constraints.
Innovation Solution
An on-device incremental image clustering system that uses both face and body features, incorporates timestamp and GPS metadata, and employs a delayed clustering mechanism to handle continuously growing galleries, ensuring high precision and recall.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only face appearance features are used for pet recognition, then precision for pet identification is improved, but images where only pet bodies are visible are ignored
Solution Approach 1:
The patent combines face appearance features and body appearance features into a unified feature representation. Face features are extracted using a face recognition model, while body features are extracted using a body recognition model. Both feature types are integrated to form comprehensive pet representations that enable accurate recognition whether the pet's face or body is visible in the image.
Solution Approach 2:
The system implements multiple recognition models (face recognition model and body recognition model) that can handle different types of pet images. The face recognition model processes images with visible pet faces, while the body recognition model processes images with only pet bodies visible, making the system universally applicable to various pet image scenarios.
2Reliability
If batch mode clustering is used, then processing completeness is improved, but the system cannot handle continuously growing galleries efficiently
Solution Approach 1:
The patent implements an incremental clustering approach where the clustering structure is dynamically updated as new images are added to the gallery. Instead of re-clustering all images from scratch, the system incrementally integrates new images into existing clusters by comparing their features with cluster centroids, making the system adaptable to continuously growing galleries.
Solution Approach 2:
The system pre-computes and stores cluster centroids for each pet identity based on existing images. When new images are added, the system compares their features against these pre-computed centroids to quickly determine cluster assignments, enabling efficient processing without requiring complete re-clustering.
3Power
If cloud-based infrastructure is used, then processing power is improved, but user privacy and connectivity constraints are not respected
Solution Approach 1:
The patent implements on-device processing where the mobile device itself performs feature extraction, comparison, and clustering operations using its own computational resources. The device stores pet image data locally and executes the clustering algorithm without requiring cloud connectivity, enabling the system to function autonomously while respecting user privacy and offline constraints.
Data Source
AI summary
A method performed by at least one processor includes receiving a plurality of images; detecting an object in at least one image from the plurality of images; performing feature extraction on the object to extract a first feature of the object and extract a second feature of the object; selecting an image from the plurality of images; based on determining the selected image includes the first feature, adding the selected image to a cluster associated with the object; and based on determining the selected image does not include the first feature and includes the second feature, adding the selected image to the cluster associated with the object based on determining that the second feature satisfies a feature distance condition.