Clustering Recognition Positions for Image Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for image recognition require extensive human labeling of object positions or orientations in images, which is labor-intensive and costly due to the need to correct numerous recognition results.
Innovation Solution
An image processing apparatus that uses clustering to obtain representative positions or orientations from recognition results, allowing users to edit and save these as learning data for improving recognition accuracy, thereby reducing the load on human correctors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the recognizer recognizes the position or position/orientation of the target object, then the recognition result is obtained, but the cost to correct each recognizer result by human is high due to enormous number of pixels
Solution Approach 1:
The patent segments the enormous number of pixel-level recognition results into a smaller number of object-level results. By performing clustering on recognition positions and extracting representative positions, the system divides the correction task from individual pixels to consolidated object representations, making human correction feasible and efficient.
Solution Approach 2:
The patent extracts representative positions from the large set of recognition positions through clustering. This extraction process identifies key positions that represent groups of similar recognition results, allowing users to correct only these representative positions rather than all individual pixel-level results.
2Reliability
If human labeling is performed on color image or depth image to learn feature or pattern, then the target object recognition is enabled, but it is difficult for human to correctly perform whole labeling due to enormous amount of data required
Solution Approach 1:
The system performs self-service by automatically generating initial recognition results and clustering them into representative positions. This self-processing reduces the burden on human labelers, who only need to review and correct the already-processed representative positions rather than performing complete labeling from scratch.
Solution Approach 2:
The system implements a feedback mechanism where recognition results are continuously improved through clustering and representative position extraction. The corrected representative positions feed back into the learning process, enabling iterative refinement of recognition accuracy while reducing the complexity of manual labeling.
Data Source
AI summary
A plurality of recognition positions each recognized by a recognizer as a position of a target object on an input image are acquired. At least one representative position is obtained by performing clustering for the plurality of recognition positions. The representative position is edited in accordance with an editing instruction from a user for the representative position. The input image and the representative position are saved as learning data to be used for learning of the recognizer.