Product Image Similarity Analysis for Obscured Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image analytical techniques require significant computational resources, are cumbersome to implement, and struggle with identifying objects obscured by obstructions, especially in moving images, leading to inaccurate object identification and slow integration with downstream systems.
Innovation Solution
A two-phase image analysis process utilizing a detection model to identify object locations and classifications, followed by a similarity model to generate embeddings for product identification, reducing resource intensity and improving accuracy by filtering irrelevant products.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image analytical techniques are used, then object identification can be performed, but computational resources are significantly consumed and processing time increases
Solution Approach 1:
The patent segments the image analysis process into distinct phases: detection phase for locating objects and similarity phase for identifying obscured objects. This segmentation allows each phase to use optimized techniques appropriate to its specific task, improving overall efficiency while maintaining accuracy.
Solution Approach 2:
The detection model performs preliminary action by identifying and locating objects in the image before the similarity analysis phase. This preliminary object location information is used to guide subsequent similarity comparisons, reducing the computational scope and resources required for full image analysis.
2Measurement precision
If conventional image analytical techniques are used, then objects can be identified, but implementation becomes cumbersome and integration with downstream systems is slow
Solution Approach 1:
The system is segmented into modular components: image sensor, detection model, similarity model, and downstream system interfaces. Each module performs a specific function and can be independently optimized or replaced, reducing overall system complexity while maintaining identification capabilities.
Solution Approach 2:
The patent introduces intermediate data structures and processing layers between the image input and downstream systems. These intermediaries standardize data formats and processing workflows, making integration with various downstream systems easier and reducing implementation complexity.
3Productivity
If conventional image analytical techniques are used, then object detection can be performed, but accuracy decreases when obstructions are present
Solution Approach 1:
The similarity model acts as an intermediary that compares obscured objects with reference images from a database. This intermediary process enables accurate identification of obscured objects by matching partial or blocked views against complete reference images, overcoming the limitations of direct detection methods.
Solution Approach 2:
A database of reference images is prepared in advance through preliminary action. These pre-stored reference images enable the similarity model to quickly and accurately identify obscured objects by comparison, improving accuracy without adding computational overhead during real-time processing.
4Measurement precision
If large data sets are used for training, then model accuracy improves, but computational resources and processing time increase
Solution Approach 1:
The patent extracts and stores only the essential features and characteristics of products in a structured database format. This extraction process creates a compact representation that maintains model accuracy while significantly reducing the computational resources required for processing compared to using complete image datasets.
Solution Approach 2:
Product data and reference images are prepared and organized in advance through preliminary action. This pre-processing creates optimized data structures that can be efficiently queried and compared during runtime, reducing computational resource consumption during actual object identification while maintaining high accuracy.
Data Source
AI summary
A computer-implemented method for product identification and classification in an image includes receiving, with one or more processors, an image containing a being a product and inputting the received image to at least one model. The at least one model may be configured to: identify a location of the product within the image, output the location of the product within the image as a crop image, generate a product classification for the product in the crop image, and generate a product embedding according to the product classification. The method may further include outputting the product embedding to a search service configured to return product data associated with at least one similar product, receiving, with the one or more processors, the product data returned by the search service, and generating, with the one or more processors, one or more image tags based on the at least one similar product.