Image Subject Area Recognition via Candidate Extraction and Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing methods face challenges in accurately and efficiently detecting subject areas in images, particularly when multiple subjects are present, leading to unstable analysis results, high computational complexity, and long processing times.
Innovation Solution
A method and apparatus that extract candidate subject areas from an image, eliminate non-subject areas, adjust and classify the remaining areas, and merge them to precisely recognize the subject areas, utilizing a combination of common subject detection models, cascade classifiers, and machine learning algorithms for efficient and accurate identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection way is used to detect subject areas in images with multiple subjects, then the detection capability is improved, but the calculation complexity increases and processing time is extended
Solution Approach 1:
The patent segments the image processing task into multiple stages: first extracting candidate subject areas using a detection model, then filtering non-subject areas through classification, and finally merging results. This multi-stage segmentation reduces the computational burden on any single algorithm while maintaining high detection accuracy for images with multiple subjects.
Solution Approach 2:
The patent performs preliminary extraction of candidate subject areas before final detection and classification. By pre-identifying potential subject regions using a detection model, the system narrows down the search space for subsequent classification and merging operations, thereby reducing overall calculation complexity while preserving detection precision.
2Productivity
If deep neural network is used for subject coordinate regression, then the detection speed is improved, but the detection accuracy for multiple subjects deteriorates and only one subject area is returned
Solution Approach 1:
The patent divides the detection process into candidate extraction, classification filtering, and result merging stages. The deep neural network is used in the first stage for fast candidate generation, while subsequent stages handle multi-subject precision through classification and intelligent merging, thus achieving both speed and accuracy.
Solution Approach 2:
The patent merges multiple candidate subject areas after classification filtering to produce the final detection result. This merging step combines the speed advantage of deep neural network regression with the accuracy of classification-based filtering, enabling accurate detection of multiple subjects while maintaining efficient processing.
3Ease of manufacture
If significant energy function is used for subject area detection, then the processing simplicity is improved, but the detection accuracy and validity deteriorate
Solution Approach 1:
The patent segments the detection process into multiple specialized stages: candidate extraction using detection models, filtering using classification algorithms, and final merging. This segmentation replaces the single-step significant energy function approach with a multi-stage process that achieves higher accuracy while maintaining reasonable processing simplicity through modular design.
Solution Approach 2:
The patent introduces classification algorithms as an intermediary between simple energy-based detection and final subject area determination. This intermediary step filters out non-subject areas from candidate regions, significantly improving detection accuracy and validity while adding only moderate computational complexity.
Data Source
AI summary
A method and a device for recognizing a subject area of an image are provided. The method includes receiving an image to be recognized, and extracting a candidate subject area from the image; eliminating a non-subject area from the extracted candidate subject areas to obtain a target candidate subject area, and adjusting and classifying the target candidate subject areas to obtain at least one set of the target candidate subject areas; and merging the target candidate subject areas in each set of the target candidate subject areas to recognize the subject area of the image.