摘要
真实路况中的运动车辆图像进行图像分割时,图像中往往存在多个车辆车牌信息,且这些车牌信息具有尺度不一,位置随机等特点,加之光照及复杂背景的影响,如何兼顾多个车辆车牌的分割效果是车辆检测和跟踪领域亟待解决的问题.为了解决这类工程应用中的问题,需要在尺度空间下对多目标图像进行分析.因本文在前期多尺度分割模型的基础上引入视觉注意机制,利用不变性特征实现多目标的定位及最优分割尺度的选取.经大量实验测试结果表明,该算法较好地实现了图像中多个车牌图像的分割并且具有较好的分割效果.
When the moving vehicle image under the real road condition is segmented,the multiple license plates have random positions and different scales.Under the influence of illumination and complex background,how to give considerations to the segmentation results for several objects in an image is the problem urgently to be resolved in the vehicle detection and tracking field.In order to solve such problem in engineering application,multiple targets image should be analyzed in the scale space.Therefore,the visual attention mechanism is introduced on the basis of the RF-PCNN model,using invariant feature to realize the location for the multiple targets and select the optimal segmentation scale.Thus all the license plates can be segmented in the vehicle image.A large number of experimental results show that the algorithm has implemented the multiple targets segmentation and has better segmentation effect.
出处
《北京交通大学学报》
CSCD
北大核心
2014年第3期118-122,共5页
JOURNAL OF BEIJING JIAOTONG UNIVERSITY
基金
国家自然科学基金资助项目(61227003)
国家科技重大专项资金资助(2012ZX07205-005-07)
广东省战略性新兴产业发展专项资金支持项目(2011912030)
关键词
图像分割
视觉注意机制
尺度空间
不变性特征
复杂背景
多目标图像
image segmentation
visual attention mechanism
scale space
invariant feature
complex background
multiple targets image
作者简介
杨娜（1977-）,女,山西长治人,讲师,博士.研究方向为交通图像处理.email：08111010@bjtu.edu.cn.
