Fuzzy license plate reconstruction method based on characteristic learning
A feature learning and blurring technology, applied in the field of image processing, can solve problems such as motion blur, inability to restore license plate images, defocus blur, etc., and achieve good reconstruction effects, improved recognizability, and accurate restoration effects
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0093] refer to figure 1 , a fuzzy license plate reconstruction method based on feature learning, including:
[0094] S1. Obtain a large number of clear character samples covering all license plate character types and obtain their corresponding fuzzy character samples. After establishing the fuzzy character sample library, perform feature training on the fuzzy character sample library, specifically:
[0095] After obtaining a large number of clear character samples covering all license plate character types, each clear character sample is convolved with various degraded functions that simulate realistic scenes to generate corresponding fuzzy character samples, and then all fuzzy character samples are established after the fuzzy character sample library , to perform feature training on the fuzzy character sample library. Wherein, the step of carrying out feature training to the fuzzy character sample database is as follows: according to the following formula, each class of fuz...
Embodiment 2
[0120] This embodiment is basically similar to Embodiment 1, and the difference is that in step S4, what this embodiment adopts to the segmented characters is a block classification method, and step S4 includes:
[0121] S41. Refer to figure 2 As shown, each segmentation character is divided into six uniform small blocks, and each small block is matched with the corresponding small block of each type of fuzzy character samples one by one to obtain the most likely top three classifications; specific to 34 types license plate character type, each small block is matched one by one with the training results of the small blocks of 34 corresponding positions. Multiply with the feature matrix (training result) of the 34 corresponding small blocks, and then classify according to the result sparsity, and use the three result vectors with the highest sparsity as the first three classifications of the small block.
[0122] S42. According to the prior condition of connectivity between a...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com