Image Recognition Apparatus Dual-Threshold Object Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems face challenges in accurately determining whether a detected object in an image is not registered in a database, particularly when the object's status, such as orientation or lighting conditions, differs from the registered data.
Innovation Solution
The system employs a dual-threshold approach, where a recognition evaluation value is compared against a first threshold to determine if the object is registered, and if not, a status coincidence degree is evaluated against a second threshold to confirm the object's difference from the database entries, thereby enhancing the accuracy of non-registration determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single threshold is used for similarity comparison, then the recognition process is simple and fast, but the accuracy of determining whether an object is registered in the database deteriorates when object status changes
Solution Approach 1:
The patent segments the recognition evaluation into two distinct components: similarity degree (feature matching) and status coincidence degree (condition matching). This segmentation allows the system to independently evaluate both the object's features and its status conditions, resolving the contradiction by maintaining simple threshold comparison while improving recognition accuracy through multi-dimensional evaluation.
Solution Approach 2:
The patent adds a new evaluation dimension (status coincidence) to the traditional single-dimension similarity comparison. By introducing status as a separate evaluation axis with its own threshold, the system transforms the recognition problem from one-dimensional to two-dimensional, enabling accurate determination of object registration status even when object conditions change.
2Productivity
If similarity comparison is performed without considering object status, then the processing is fast and simple, but false positives occur when the same object appears under different conditions
Solution Approach 1:
The patent divides the recognition evaluation into separate similarity assessment and status assessment components. This segmentation allows the system to maintain fast processing through independent threshold comparisons while improving reliability by ensuring both feature matching and status consistency are satisfied for accurate recognition.
Solution Approach 2:
The patent introduces status coincidence degree as an intermediary evaluation criterion between the object image and database entries. This intermediary layer acts as a filter that prevents false positives by verifying that not only do features match but also that the object's status conditions align, thereby improving recognition reliability without significantly impacting processing speed.
3Ease of manufacture
If feature amounts are compared directly without status normalization, then the comparison is straightforward, but the similarity value becomes unreliable when lighting or orientation changes
Solution Approach 1:
The patent segments the feature comparison process into feature extraction (simple) and status evaluation (condition-based) components. This segmentation maintains implementation simplicity by keeping the core feature comparison straightforward while adding status as a separate, easily evaluable criterion that corrects for lighting and orientation variations.
Solution Approach 2:
The patent changes the evaluation parameters from raw feature amounts alone to a composite of similarity degree and status coincidence degree. By introducing status parameters (lighting conditions, orientation) as separate evaluation dimensions, the system maintains simple implementation while significantly improving measurement accuracy under varying conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image recognition apparatus detects a specific object image from an image to be processed, calculates a coincidence degree between an object recognisability state of the object image and that of an object in registered image information, and calculates a similarity between the image feature of the object image and the image feature in the registered image information. Based on the similarity and coincidence degree, the image recognition apparatus recognizes whether the object of the object image is that of the registered image information. When the similarity is lower than the first threshold and the coincidence degree is equal to or higher than the second threshold, the image recognition apparatus recognizes that the object of the object image is different from that of the registered image information.