基于低分辨率车牌识别系统的应用研究

Research on Application of Low-Resolution License Plate Recognition System

摘要: 在自然场景下的监控视频中，由于拍摄距离和拍摄角度造成的车牌图像有效分辨率低问题，超分辨率(Super-Resolution, SR)技术可有效解决这个问题，然而还有很多车牌图像存在分辨率过低、模糊及光照问题，对于这些图像做超分辨率后仍存在无法识别的现象，且会占用计算资源和降低速度。针对该类问题，提出一种无参考图像质量评估系统以甄别出有效图像进行超分辩率和车牌识别。根据提出的场景需求，归纳出四种车牌图像质量评价因子：分辨率、清晰度、亮度及对比度，最后使用支持向量机(Support Vector Regression, SVR)来进行质量分数回归预测。实验表明本文算法在保证速度的同时可有效提高车牌识别准确率。

Abstract: In the surveillance video in the natural scene, Super-Resolution (SR) technology can effectively solve this problem due to the low effective resolution of the license plate image caused by the shooting distance and shooting angle. However, there are still many license plate images with low resolution, blur and illumination problems. There is still an unrecognizable phenomenon after super-resolution of these images, which will occupy computing resources and reduce the speed. Aiming at this kind of problem, a non-reference image quality evaluation system is proposed to identify the effective image for super-variation and license plate recognition. According to the scene requirements presented, four license plate image quality evaluation factors are summarized: resolution, sharpness, brightness and contrast. Finally, support Vector Regression (SVR) is used to perform mass fraction regression prediction. Experiments show that the proposed algorithm can effectively improve the accuracy of license plate recognition while ensuring the speed.