Image Processing Apparatus for Pose Search via Feature Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies face challenges in accurately searching for images depicting the same pose or action due to physical and individual differences, leading to difficulties in identifying similar images.
Innovation Solution
An image processing apparatus and method that acquire a query moving image and compute feature values of keypoints detected in the image. The apparatus determines correspondence relationships between frame images and integrates feature values to generate an integrated query, enabling accurate search for images with similar poses or actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single query image is used for searching, then search speed is fast, but search accuracy is low due to physical and individual differences
Solution Approach 1:
The patent merges multiple query images (first query moving image and second query moving image) into a single integrated query by computing feature values from each, determining correspondence relationships between frames, and integrating the feature values. This combined query representation captures more comprehensive pose and action information, improving search accuracy while accounting for physical and individual differences that a single image would miss.
2Measurement precision
If multiple query images are integrated, then search accuracy improves, but processing complexity increases
Solution Approach 1:
The patent segments the processing into distinct stages: extracting feature values from multiple query images separately, determining correspondence relationships between specific frames, and then integrating the feature values. This segmentation allows the system to handle complex multi-image processing systematically, managing the computational complexity through structured step-by-step operations rather than attempting to process all images simultaneously.
Solution Approach 2:
The patent introduces an intermediary correspondence relationship determination step that matches frames between the first and second query moving images based on temporal and spatial relationships. This intermediary process establishes reliable pairings before feature value integration, acting as a mediator that simplifies the overall integration task by pre-organizing which features should be combined, thereby reducing processing complexity.
3Reliability
If feature values are integrated from multiple sources, then search reliability improves, but computation time increases
Solution Approach 1:
The patent performs preliminary actions by extracting feature values from all query images and determining correspondence relationships between frames before the actual search operation. By pre-processing and organizing the feature values and their relationships in advance, the system reduces the computational burden during the search phase, thereby minimizing computation time while maintaining high reliability through comprehensive feature integration.
Data Source
AI summary
A query acquisition unit (109) acquires a first query moving image having a plurality of time-series first frame images and at least one second query moving image having a plurality of time-series second frame images. A feature value computing unit (103) computes a feature value of a keypoint detected from an object being included in each of the first frame images and the second frame images. A correspondence relationship determination unit (114) determines a second frame image associated with each of the plurality of first frame images. An integration unit (113) performs, for a plurality of combinations of the first frame image and the second frame image associated with each other, processing of integrating the feature values computed from each of the first frame image and the second frame image, and thereby generates an integrated query in which the integrated feature values are arranged in time series.