Media Clustering via Neural Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
User devices face memory inefficiencies due to redundant media content, such as duplicate images and videos, which obstruct primary functionalities and increase latency by occupying significant storage space.
Innovation Solution
A neural network model is used to extract spatial and global features from media, generating feature vectors that allow for dynamic clustering based on similarity thresholds, enabling the identification and grouping of duplicate or similar media files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If media content is stored in user device memory, then the available content increases, but redundant content occupies significant storage space and obstructs primary functionalities
Solution Approach 1:
The patent extracts feature representations from media content using neural networks, separating the essential identifying characteristics from the actual media files. This allows the system to identify and manage redundant content based on extracted features without moving or deleting the original media files, thus removing the harmful effect of redundancy while preserving the content.
Solution Approach 2:
The system identifies duplicate and near-duplicate media files through feature comparison and enables users to selectively discard redundant copies while recovering or preserving the original content. This principle allows the system to eliminate storage waste from redundant content while maintaining access to unique media files.
2Quantity of substance
If a large quantity of media content is stored, then the content variety increases, but the task of organizing data becomes burdensome
Solution Approach 1:
The patent implements self-service through automated feature extraction, similarity computation, and clustering algorithms that automatically organize media content without user intervention. The system performs duplicate detection, near-duplicate identification, and grouping based on visual and audio features, enabling automatic data organization that reduces the burden on users while managing large quantities of media content.
Solution Approach 2:
The system changes organizational parameters by using feature space distance metrics (cosine similarity, Euclidean distance) to automatically group media files. By transforming the organization task from manual file management to automated feature-based clustering with adjustable similarity thresholds, the system makes data organization easier while handling large content volumes.
3Quantity of substance
If redundant content accumulates in memory, then storage capacity is consumed, but the latency of instruction execution increases
Solution Approach 1:
The patent extracts essential feature representations from media files, creating compact feature vectors that capture the essence of each file. This extraction process enables the system to identify and manage redundant content based on these compressed representations, reducing the computational overhead of processing and accessing large volumes of media content, thereby decreasing instruction execution latency while maintaining storage capacity.
Solution Approach 2:
The system performs preliminary feature extraction and indexing of media content, preparing the data structure in advance for efficient querying and duplicate detection. By pre-processing media files to extract and store feature representations, the system reduces the time required for subsequent operations such as duplicate identification and content management, thus reducing execution latency without sacrificing storage capacity.
Data Source
AI summary
Methods and systems for grouping of media based on similarities between features of media data are provided. A method of managing a plurality of images may include: identifying a threshold distance for clustering the plurality of images based on a degree of similarity between the plurality of images; extracting a plurality of feature vectors corresponding to the plurality of images; and generating at least one cluster comprising at least two images among the plurality of images, based on cosine distances between feature vectors corresponding to the at least two images being less than the threshold distance.