Face Cluster Assignment Using Probability Databases
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video content analysis methods, particularly those relying on language analysis, struggle to effectively extract semantic information from non-English content and may not identify faces in images where the person's head is turned or obscured, leading to inefficiencies in metadata generation and face clustering.
Innovation Solution
A computer-implemented method and system that analyzes digital images to identify faces using identification regions, merges face clusters based on similarities, and assigns images to clusters using a probability-based approach, even when faces are partially obscured, by utilizing a library of face probabilities built from prior video analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If language analysis is used to generate metadata, then the system can process video content efficiently, but it fails to extract semantic information from non-English content and images with obscured faces
Solution Approach 1:
The system segments video content into individual digital images and further segments each image into multiple identification regions, allowing different processing approaches for different parts of the content. This enables the image analysis system to focus on specific regions containing faces or semantic information while maintaining overall processing efficiency.
Solution Approach 2:
The system introduces an intermediary probability database that stores pre-calculated face identification probabilities from multiple video sources. This intermediary structure bridges the gap between limited image data and accurate face identification, allowing the system to make informed decisions even when faces are obscured or partially visible.
2Loss of information
If image analysis is used to identify faces, then semantic information can be extracted from non-English content, but the system cannot identify faces when heads are turned or faces are obscured
Solution Approach 1:
The system performs preliminary analysis by building a probability database from multiple video sources before actual face identification. This pre-computation of face co-occurrence probabilities enables the system to handle obscured or turned faces by referencing previously established patterns, rather than relying solely on the current image quality.
Solution Approach 2:
The system uses feedback from the probability database to guide face identification decisions. When image analysis alone is insufficient, the system references probability information about which faces typically appear together in similar contexts, allowing it to correctly identify faces even when visual information is limited or ambiguous.
3Adaptability or versatility
If traditional face clustering is used, then faces can be grouped by similarity, but the system cannot handle faces with different poses or obscured features
Solution Approach 1:
The system adds a probabilistic dimension to traditional face clustering by incorporating face co-occurrence probabilities from the database. This transforms the clustering problem from relying solely on visual similarity to a multi-dimensional approach that combines visual features with statistical probability information, enabling accurate clustering even when visual features vary due to pose or obscuration.
Data Source
AI summary
A method includes accessing a digital image including a plurality of faces including a first face and a second face. The method includes identifying a plurality of identification regions of the digital image including a first identification region associated with the first face and a second identification region associated with the second face. The method also includes assigning the digital image to a first face cluster of a plurality of face clusters when a difference between data descriptive of the first identification region and data descriptive of a face cluster identification region of the first face cluster satisfies a threshold. The method further includes assigning the digital image to a second face cluster of the plurality of face clusters based at least partially on a frequency of the second face and the first face appearing together in an image.