Text-to-Image Matching with Wildcards for Privacy-Preserving Vehicle Re-Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current text recognition in images, particularly for license plates, faces challenges in accuracy and privacy concerns, with existing Automatic License Plate Recognition (ALPR) systems exhibiting limited re-identification accuracy and revealing personal information, and lacking the ability to handle text queries with wildcard symbols efficiently.
Innovation Solution
A method and system for text-to-image matching that embeds character strings and images into a common vectorial space using Spatial Pyramid Bag-Of-Characters (SPBOC) and Structured Support Vector Machine (SSVM) framework, enabling efficient recognition and privacy-preserving vehicle re-identification without explicit character detection, and allowing queries with wildcard characters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCR recognition is used for license plate recognition, then text can be recognized, but computational cost increases and accuracy diminishes when visibility is poor
Solution Approach 1:
The patent replaces traditional OCR mechanical character recognition with a deep learning-based image matching system. Instead of detecting and recognizing individual characters through OCR processes, the system extracts features from the entire license plate image and compares them directly with database images using neural networks, thereby reducing computational complexity while improving accuracy
Solution Approach 2:
The patent creates a universal image signature representation that can match license plates regardless of lighting conditions, angles, or partial occlusions. The feature extraction and matching system serves multiple functions: it handles both clear and poor visibility conditions, works with various license plate formats, and provides robust matching across different environmental conditions
2Measurement precision
If exact ALPR string matching is used for vehicle re-identification, then personal information is revealed, but privacy is compromised
Solution Approach 1:
The patent extracts and removes the need for explicit character recognition and text extraction from the license plate image. By using direct image-to-image matching through feature signatures, the system eliminates the step that would reveal readable text information, thereby protecting privacy while maintaining re-identification accuracy
Solution Approach 2:
The patent introduces image signatures and feature vectors as intermediaries between the license plate image and the vehicle identification process. Instead of directly reading and processing text characters, the system uses these intermediate feature representations that capture the visual essence of the license plate without exposing readable information, thus mediating between recognition accuracy and privacy protection
3Adaptability or versatility
If database of possible license plate numbers is maintained for ALPR queries, then wildcard queries can be handled, but system complexity and storage requirements increase
Solution Approach 1:
The patent replaces the traditional approach of maintaining exhaustive databases of possible license plate numbers with deep learning-based image matching. The neural network system inherently handles wildcard queries by comparing image features directly, eliminating the need for complex database searches and reducing system complexity while maintaining versatility
Solution Approach 2:
The patent changes the fundamental parameter of query processing from text-based string matching with wildcard expansion to image-based feature matching. This parameter change allows the system to handle wildcard queries naturally through visual similarity comparison, reducing the need for maintaining large databases of possible combinations and simplifying the overall system architecture
Data Source
AI summary
A system and method for comparing a text image with or without a wildcard character and a character string are provided. The method includes embedding a character string into a vectorial space by extracting a set of features from the character string and generating a character string representation based on the extracted features, such as a spatial pyramid bag of characters (SPBOC) representation. A text image is embedded into a vectorial space by extracting a set of features from the text image and generating a text image representation based on the text image extracted features. A similarity between the text image representation and the character string representation is computed, which includes computing a function of the text image representation and character string representation.