Retail Image Recognition via Dual-Resolution Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Object recognition in retail environments is challenging due to high variability in product appearance attributes and frequent introduction of new products, which current systems struggle to adapt to effectively.
Innovation Solution
A method and system that utilize an imaging device to capture and process retail images, transmitting low-resolution images for real-time recognition and high-resolution images for updating visual signatures, enabling accurate identification and enrichment of the item database with new products.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution images are used for all products, then recognition accuracy is improved, but data transmission time and storage requirements increase significantly
Solution Approach 1:
The patent segments the image processing workflow into two distinct pathways: a fast recognition pathway using low-resolution images for common products, and a precise identification pathway using high-resolution images for new or unrecognized products. This segmentation allows the system to achieve high recognition accuracy only when necessary, while maintaining fast response times for the majority of products.
Solution Approach 2:
The system applies high-resolution image processing only partially - specifically for new products or products that cannot be recognized with low-resolution images. For all other products, low-resolution images are sufficient. This partial application of high-resolution processing optimizes the balance between recognition accuracy and transmission time.
2Adaptability or versatility
If the database includes all possible product variations, then recognition coverage is improved, but system complexity and update frequency increase
Solution Approach 1:
The system enables automatic database enrichment through self-service mechanisms. When a new product is detected via high-resolution imaging, the system automatically extracts product features, generates visual signatures, and updates the database without requiring manual intervention. This self-service approach improves recognition coverage while avoiding the complexity of manual database management.
Solution Approach 2:
The system performs preliminary analysis using low-resolution images to identify potential new products before committing resources to high-resolution processing. This preliminary action filters out already-known products, ensuring that high-resolution processing and database updates are only triggered when necessary, thereby reducing overall system complexity.
3Productivity
If low-resolution images are used for processing, then transmission speed is improved, but recognition accuracy for new products decreases
Solution Approach 1:
The system dynamically adjusts image resolution based on real-time needs. Low-resolution images are used for initial processing and common products to maximize transmission speed. When a product is identified as new or unrecognized, the system dynamically switches to high-resolution imaging to ensure accurate recognition, thereby optimizing both speed and accuracy adaptively.
Solution Approach 2:
Low-resolution images serve as an intermediary step in the recognition process. They provide a quick first pass for identification, and only when this intermediary analysis fails to recognize a product does the system proceed to high-resolution processing. This intermediary approach maintains high transmission speed while preserving the capability for accurate new product recognition.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides a method of image processing comprising: obtaining by an imaging device a low resolution version and a high resolution version of a retail image, the high resolution version of the retail image being a temporary file to be erased automatically after a predetermined time period; transmitting to a server the low resolution version of the retail image; upon receipt of a request from the server, the request including data representative of a contour of an unidentified item in the low resolution version of the retail image, cropping a high resolution item image from the high resolution version of the retail image, the high resolution item image corresponding to the contour of the unidentified item; and transmitting the high resolution item image to the server thereby enabling updating an item database.