Iterative Image Localization Through Bounding-Box Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image detection methods suffer from inadequate localization accuracy of bounding boxes, often including redundant information or missing portions due to variations in angle, illumination, occlusion, and complex backgrounds, which affects detection accuracy.
Innovation Solution
An iterative localization-type image detection method using deep learning, involving image detection, deduplication, position correction, and segmentation to refine bounding boxes, ensuring high precision and recall rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image detection is performed using conventional bounding box methods, then detection speed is maintained, but localization accuracy deteriorates due to redundant information and incomplete portion detection
Solution Approach 1:
The patent segments the image detection process into multiple stages: initial bounding box detection, deduplication of overlapping boxes, position correction using coordinate refinement, and final segmentation verification. This multi-stage segmentation approach improves localization accuracy by progressively refining bounding boxes while managing complexity through structured processing steps.
Solution Approach 2:
The patent performs preliminary deduplication and position correction actions before final segmentation. By pre-processing bounding boxes to remove duplicates and correct positions early in the pipeline, the system establishes accurate localization foundations that propagate through subsequent processing stages, improving overall measurement precision without proportionally increasing complexity.
2Reliability
If multiple detection boxes are generated to ensure high recall, then detection coverage is improved, but redundancy increases and precision deteriorates
Solution Approach 1:
The patent implements feedback loops where detection results are continuously refined through deduplication algorithms that compare overlapping boxes, position correction mechanisms that adjust coordinates based on segmentation boundaries, and iterative verification against ground truth data. This feedback-driven refinement maintains high recall by keeping multiple candidate boxes while improving precision through systematic elimination and correction of redundant detections.
Solution Approach 2:
The patent dynamically changes parameters such as bounding box coordinates, confidence thresholds, and overlap ratios throughout the processing pipeline. By adjusting these parameters adaptively during deduplication and position correction stages, the system optimizes the balance between recall (maintaining multiple candidates) and precision (refining final selections) based on local image characteristics and detection confidence levels.
Data Source
AI summary
An image processing method includes, performing image detection on a portion of an image to obtain a first box set for marking an image detection result; deduplicating a first box in the first box set to obtain a second box set; performing position correction on a second box in the second box set to obtain a third box; generating, based on the third box, a third box set of adjusted boxes and a cropped image set including cropped images that are obtained by cropping the image based on the third box set; performing recognition and segmentation on the cropped images to obtain a segmentation result including segmented images, a number of the segmented images being less than or equal to a number of the adjusted boxes; and correcting the adjusted boxes, based on coordinate positions of the segmented images, to generate corrected boxes for the portion.