Facial Recognition Grouping via Two-Pass Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional facial recognition technologies and workflows in photograph organization often produce duplicative groups for the same person, complicating the task of organizing large collections of digital photos and requiring excessive manual effort to correct.
Innovation Solution
A photograph organization module employing two passes over the photographs, with the first pass detecting faces, calculating facial representations, and assigning them to groups based on similarity scores, and the second pass merging groups linked by relationship indicators to eliminate duplicative groups, using a merging operation to combine groups with sufficient matching faces into a single group.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional facial recognition technologies are used to organize photographs, then photographs can be grouped by detected faces, but duplicative groups for the same person are produced
Solution Approach 1:
The patent divides the photograph organization process into two distinct passes: a first pass that performs initial facial recognition and grouping, and a second pass that performs merging operations on the generated groups. This segmentation allows each pass to focus on specific tasks, improving overall grouping accuracy while maintaining automation.
Solution Approach 2:
The first pass performs preliminary facial recognition and creates initial groups before the second pass merges duplicative groups. This preliminary action establishes a structured foundation that enables more efficient detection and merging of duplicates in the subsequent pass.
2Reliability
If multiple groups are created for each detected face, then comprehensive coverage is achieved, but the number of groups increases excessively
Solution Approach 1:
The second pass systematically merges groups that contain duplicative faces by comparing facial features across groups and combining overlapping groups into single unified groups. This reduces the total number of groups while preserving all detected faces and their associations.
Solution Approach 2:
The system uses feedback from the first pass results to guide the second pass merging operations. By analyzing the distribution and overlap of faces across generated groups, the system intelligently determines which groups to merge, optimizing the reduction of group count while maintaining detection completeness.
3Manufacturing precision
If manual review is performed to correct duplicative groups, then grouping accuracy improves, but user time and effort increase significantly
Solution Approach 1:
The system performs self-correction by automatically detecting and merging duplicative groups in the second pass without requiring manual user intervention. The algorithm independently identifies overlapping groups and combines them, eliminating the need for users to manually review and correct grouping errors.
Solution Approach 2:
The patent replaces manual mechanical review processes with automated computational algorithms that compare facial features, identify duplicative groups, and perform merges. This substitution of human effort with automated image processing and pattern recognition significantly reduces the time and effort required while maintaining or improving accuracy.
Data Source
AI summary
Photograph organization based on facial recognition is described. In one or more embodiments, a photograph organization module obtains multiple photographs having images of faces and recognizes the faces in the multiple photographs. The module builds a population by attempting to distinguish individual persons among the faces in the multiple photographs, with each person of the population corresponding to a group of multiple groups. After a first pass through the faces, the population includes a number of duplicative persons. With a second pass, the photograph organization module reduces the number of duplicative persons in the population by merging two or more groups of the multiple groups to produce a reduced number of groups. The merging is performed based on comparisons of the faces corresponding to the two or more groups. The multiple photographs are organized based on the reduced number of groups. Organization can include tagging or displaying grouped photographs.