基于最大稳定极值区域与主要旋转不变局部二值模式的复杂场景车牌检测方法

为了解决在复杂场景中进行车牌定位的问题,提出了一种基于MSER与DRLBP特征的车牌定位方法。首先对输入图像进行预处理,然后在多个通道上进行MSER候选区域提取;接着利用所设计的基于车牌字符合并的车牌定位方法进行车牌字符合并;最后利用DRLBP纹理特征对合并后的区域进行验证从而得到最终的车牌区域。实验结果表明该方法具有较好的定位能力。由于方法是通过寻找车牌字符进而定位车牌位置,因此其受车牌颜色、车牌格式的影响较小,在复杂环境中对国内外不同车牌均有较好定位效果。

A Method of License Plate Location in Complex Scene Based on Maximally Stable Extremal Regions and Dominant Rotated Local Binary Patterns

In order to solve the difficulty that locating vehicle license plate in a complex scene, this paper proposed a new method based on MSER algorithm and DRLBP texture feature. Firstly, preprocessing an input image and then using MSER algorithm to extract candidate license plate character regions in multiple channels. Secondly, implementing a new method of license plate location based on merging license plate characters, and use it to roughly locate the license plate region. Finally, utilizing DRLBP feature to relocate the license plate in the preceding result to produce the final detected license plate region. The experimental results show that this method is very effective. Due to this method starts from the license plate characters, and finding the license plate area by merging license plate characters. It is not only less affected by the color and format, but also effective for domestic and foreign license plate.