Method and device for generating complex license plates and training license plate generation model
By combining global text control conditions and multi-level local control conditions with random noise feature fusion, high-quality complex license plate samples are generated. This solves the problems of insufficient efficiency and quality in generating license plate images in existing technologies, reduces the training cost of license plate recognition models, and improves the generalization ability of the models.
CN122116020APending Publication Date: 2026-05-29JINAN BOGUAN INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JINAN BOGUAN INTELLIGENT TECH CO LTD
- Filing Date
- 2024-11-29
- Publication Date
- 2026-05-29
AI Technical Summary
Technical Problem
Existing license plate recognition models rely on a large number of labeled sample license plate images for training. High-quality sample license plate images covering multiple scenarios are difficult to generate efficiently, leading to increased R&D costs.
Method used
By acquiring global text control conditions and multi-level local control conditions, and combining them with random input noise, a pre-trained license plate generation model is used to perform feature fusion, generating high-quality complex license plate samples.
Benefits of technology
This improved the efficiency and quality of generating sample license plate images required for license plate recognition model training, reduced data collection costs, and enhanced the model's generalization ability.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122116020A_ABST
Abstract
The application discloses a complex license plate generation method and a license plate generation model training method and device. The complex license plate generation method comprises the following steps: obtaining a global text control condition of a to-be-generated license plate, and determining a multi-level local control condition according to the global text control condition; inputting the global text control condition, the multi-level local control condition and random input noise into a pre-trained license plate generation model, performing feature fusion on the global text control condition, the multi-level local control condition and the random input noise through the license plate generation model, and obtaining predicted noise with fused license plate features; and obtaining an output of the license plate generation model according to the predicted noise and the random input noise, and taking the output as a license plate generation result. The application increases the multi-level local control condition on the basis of the global text control condition, realizes the local level control on the basis of the global control, enhances the accuracy of generating the license plate content of each level, and improves the quality of the generated overall license plate.
Need to check novelty before this filing date? Find Prior Art