Object Recognition via Saliency Scoring and Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition methods based on strongly supervised training samples are time-consuming and laborious, as they require retraining for newly obtained data and categories.
Innovation Solution
An object recognition method that decouples body region detection from object recognition, using a saliency score to determine if an object in a body region should be recognized, and employing a feature vector and center feature vector to efficiently determine object categories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional supervised training methods are used for object recognition, then recognition accuracy for predefined categories can be achieved, but retraining is required for new categories which is time-consuming and laborious
Solution Approach 1:
The patent segments the object recognition process into two independent parts: body region detection (using saliency scores) and object recognition (using feature vectors and center feature vectors). This segmentation allows the recognition component to be updated without retraining the entire system, resolving the contradiction between maintaining accuracy and reducing retraining time.
Solution Approach 2:
The patent pre-establishes a feature library containing center feature vectors for multiple categories before actual recognition tasks. This preliminary preparation enables rapid recognition of new categories without requiring time-consuming retraining, as the system can directly compare extracted features against the pre-built library.
2Reliability
If feature vectors of all objects in the feature library are compared to determine category, then comprehensive recognition can be achieved, but recognition efficiency is reduced
Solution Approach 1:
The patent extracts only the essential center feature vectors representing each category from the feature library, rather than storing and comparing all individual object feature vectors. This extraction maintains recognition comprehensiveness while dramatically improving efficiency by reducing the comparison space to one representative vector per category.
Solution Approach 2:
The patent changes the recognition approach from comparing against multiple individual feature vectors to comparing against a single center feature vector per category. This parameter change (from many vectors to one representative vector) maintains the ability to comprehensively recognize categories while significantly improving recognition efficiency.
3Device complexity
If body region detection and object recognition are coupled, then integrated processing can be achieved, but flexibility for updating categories and feature vectors is reduced
Solution Approach 1:
The patent explicitly segments body region detection from object recognition into two independent modules. The body region detection module identifies potential object locations using saliency scores, while the recognition module independently performs category classification using feature vectors. This segmentation provides flexibility to update categories and feature vectors without affecting the detection module.
Solution Approach 2:
The patent creates a dynamic system where the feature library and category definitions can be updated independently of the detection mechanism. The decoupled architecture allows the recognition component to adapt to new categories by simply updating the feature library, while the body region detection remains unchanged and reusable.
Data Source
AI summary
This application discloses an object recognition method and apparatus in the field of artificial intelligence. This application relates to the field of artificial intelligence, and specifically, to the field of computer vision. The method includes: obtaining one or more body regions of a to-be-recognized image; determining a saliency score of each of the one or more body regions; and when a saliency score of a body region A is greater than or equal to a categorization threshold, determining a feature vector of an object in the body region A based on a feature of the object in the body region A, and determining a category of the object in the body region A based on the feature vector of the object in the body region A and a category feature vector in a feature library, where the body region A is any one of the one or more body regions.