Cognitive Visual Classification of Similar Planar Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition systems in retail and industrial settings face challenges in accurately detecting objects from image data, especially when dealing with large datasets and subtle differences between candidates, relying on expensive and time-consuming human-labeled training sets.
Innovation Solution
An automatic object recognition system that iteratively processes image data, aligns candidate templates with query images, selects points of highest saliency, and compares similarity to narrow down candidates, potentially using K-means clustering, to achieve high accuracy in classifying planar objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional supervised learning is used for object recognition, then the system can be trained with labeled images, but the process becomes expensive and time-consuming due to requiring extensive human input for labeling
Solution Approach 1:
The system performs self-labeling by automatically generating candidate templates and hypotheses from the image data itself, eliminating the need for external human annotators. The cognitive process autonomously identifies objects and creates training data through its own analysis capabilities.
Solution Approach 2:
The patent introduces an intermediary cognitive process that acts as a bridge between raw image data and final recognition results. This intermediary layer generates candidate templates and hypotheses, reducing the need for direct human involvement in the labeling process.
2Adaptability or versatility
If conventional supervised learning is used for object recognition, then the system can classify objects, but it cannot scale up to handle very large data sets comprising hundreds of thousands of objects and millions of images
Solution Approach 1:
The patent segments the large-scale object recognition task into manageable components: generating candidate templates, creating hypotheses, measuring saliency, and iteratively refining classifications. This segmentation enables the system to handle large datasets by processing them through modular cognitive operations.
Solution Approach 2:
The system employs dynamic iterative processing where candidate templates and hypotheses are continuously refined through multiple cognitive operations. This dynamic approach allows the system to adapt to large datasets by progressively improving classifications through repeated measurement and comparison cycles.
3Productivity
If automated recognition is implemented for objects with subtle differences, then the system can process multiple objects, but the list of hypotheses becomes very long and classification accuracy decreases
Solution Approach 1:
The patent applies local quality by measuring saliency at specific selected points rather than uniformly across entire images. This allows the system to focus computational resources on discriminative local features that distinguish between similar objects, maintaining accuracy while processing many candidates.
Solution Approach 2:
The system changes parameters iteratively by adjusting the set of candidate templates and hypotheses through multiple cognitive operations. By modifying the parameter set across iterations and focusing on high-saliency points, the system narrows down long hypothesis lists while maintaining or improving classification accuracy.
Data Source
AI summary
A cognitive system and method for visual classification of similar planar objects is disclosed. The method uses a query image and a list of candidate templates as the input, and produces the most probable candidate for the query image. The system uses the mutual saliency among a sample of points in the query templates, and selects those points with the highest saliency. The corresponding candidate templates to the points with the highest saliency are then compared to the query image, and those with the highest similarity are kept. The system has applications to industrial and commercial settings where processes require object recognition from image data.