基于视觉注意机制 PCNN 模型的车牌 图像分割方法

车辆图像中车牌具有所占比例小、位置不固定、大小不一，以及分割易受光照影 响的特点.因此，车牌图像的分割始终是车辆跟踪、车辆识别等领域中的难点问题.针对以 上问题，本文提出了基于视觉注意机制脉冲耦合神经网络模型的车牌图像分割方法.该方 法将视觉注意机制中的数据驱动模式和任务驱动模式相结合. 数据驱动模式中，通过对 PCNN 模型细胞感受野功能的完善，使优化 PCNN 模型具有了尺度性和方向性.任务驱 动模式中，针对不同尺度的分割，利用组合不变矩和局部灰度熵，自适应地确定目标的特 征尺度和最佳尺度，并确定该目标最终的分割结果.经实验验证，该方法对车牌图像具有 较好的分割效果.

A Segmentation Method of License Plate Image Based on PCNN Model with Visual Attention Mechanism

In the vehicle image, the license plate bares the characteristics of small proportion, random posi- tion, different sizes and sensitive to light in segmentation. The segmentation of license plate image has al- ways been one of the urgent problems in vehicle tracking and identification fields. In view of this, this paper proposes a segmentation method for license plate image based on pulse coupled neural networks model with visual attention mechanism. The method combines data driven pattern and task driven pattern of visual atten- tion mechanism. In the data driven pattern, the receptive field function of PCNN model is optimized, result- ing in scale and orientation. In task driven pattern, the target’s feature scale and optimal scale are adaptively determined using combined invariant moment and local gray entropy. The final segmentation result is pro- duced by comparing feature values at different scales. Experimental results show that the segmentation of li- cense plate image achieves good performance.