Neural Network Fingerprinting for Non-Compliant Image Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision applications are unable to efficiently identify and remove non-compliant images, such as those containing nudity, sexually explicit content, or racist content, from large collections due to storage and processing constraints, making it difficult to manage and search through vast datasets.
Innovation Solution
A system utilizing a neural network architecture with a fingerprinting network that generates feature embeddings and converts them into binary codes, allowing for efficient storage and quick searches for non-compliant content, while preserving salient feature information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review is used to identify non-compliant images, then identification accuracy is improved, but productivity deteriorates due to the inability to handle large collections
Solution Approach 1:
The patent replaces manual mechanical review with automated computer vision systems using neural networks. The system processes images through convolutional neural networks that automatically detect non-compliant content, achieving both high accuracy and scalable productivity by substituting human labor with algorithmic processing.
Solution Approach 2:
The patent introduces an intermediary indexing system that creates compact representations (hashes or embeddings) of image features. This intermediary layer enables efficient similarity search and filtering, allowing the system to rapidly identify potentially non-compliant images from large collections before applying more detailed analysis.
2Reliability
If comprehensive image analysis is performed on large collections, then identification capability is improved, but use of energy deteriorates due to storage and processing constraints
Solution Approach 1:
The patent extracts only the most salient features from images using neural network-based feature extraction. Instead of processing entire high-resolution images, the system extracts compact feature vectors or hash representations that capture essential visual characteristics, significantly reducing computational and storage requirements while maintaining identification capability.
Solution Approach 2:
The patent transforms images from their original high-dimensional pixel space into compressed feature spaces using dimensionality reduction techniques. By changing the representation parameters from millions of pixels to thousands or hundreds of feature dimensions, the system achieves efficient processing and storage while preserving the ability to identify non-compliant content.
3Measurement precision
If feature embeddings are stored for all images, then search accuracy is improved, but storage requirements deteriorate
Solution Approach 1:
The patent creates simplified copies or hash representations of image features instead of storing complete feature embeddings. These compact hashes serve as proxies that enable rapid comparison and matching while occupying minimal storage space, allowing the system to maintain search functionality across large image collections without prohibitive storage costs.
Data Source
AI summary
Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and perform: receiving images at an electronic platform; generating, using a fingerprinting network associated with a neural network architecture, feature embeddings corresponding to the images; generating, using the fingerprinting network, codes based on the feature embeddings; storing the codes in one or more databases; receiving, using a search component associated with the electronic platform, a non-compliant image comprising non-compliant content; and searching the codes in the one or more databases to identify one or more of the images that include non-compliant content similar to, or the same as, the non-compliant content included in the non-compliant image. Other embodiments are disclosed herein.