Image edge detection based on nonsubsampled contourlet transform and mathematical morphology

A novel algorithm for image edge detection is presented. This algorithm combines the nonsubsampled contourlet transform and the mathematical morphology. First, the source image is decomposed by the nonsubsampled contourlet transform into multi-scale and multi-directional subbands. Then the edges in the high-frequency and low-frequency sub-bands are respectively extracted by the dualthreshold modulus maxima method and the mathematical morphology operator. Finally, the edges from the high- frequency and low-frequency sub-bands are integrated to the edges of the source image, which are refined, and isolated points are excluded to achieve the edges of the source image. The simulation results show that the proposed algorithm can effectively suppress noise, eliminate pseudo-edges and overcome the adverse effects caused by uneven illumination to a certain extent. Compared with the traditional methods such as LoG, Sobel, and Carmy operators and the modulus maxima algorithm, the proposed method can maintain sufficient positioning accuracy and edge details, and it can also make an improvement in the completeness, smoothness and clearness of the outline.

基于模块C-2DPCA算法的人脸识别方法

提出了基于模块化完全二维主成分分析(modular C-2DPCA)算法的人脸识别方法,该方法首先对图像矩阵进行分块,将分块得到的子图像矩阵直接用于鉴别分析,由于直接基于二维子图像矩阵,能方便地降低鉴别特征的维数,在特征提取过程中可以避免使用矩阵的奇异值分解,方法简便.该方法与改进前完全二维主成分分析(C-2DPCA)方法在ORL人脸数据库上的仿真识别效果比较表明,改进后的方法在保持较高识别率的前提下鲁棒性有很大提高.

一种移动对象数据库的自适应更新方法

更新频率的选择是移动对象数据库领域的一个难题.基于局部线性化思想,给出移动对象的常速率及常加速度-常角速度动态模型.接着,模糊逻辑定量化方法被用于描述移动对象的状态及推算误差.综合考虑推算距离误差、速度变化、时间间隔等多种因素,一个"或"门计算单元被用以判断是否需要更新.利用地图匹配的计算结果,根据道路的特征,提出自适应选择动态模型的方法.最后,用跑车实验验证了所述算法性能.本算法高度可配置,能够在自适应选择更新频率的情况下,较忠实记录移动对象状态信息.

基于颜色信息定位车牌及其校正

车牌定位是车牌识别中的关键技术之一。本文对传统基于边缘检测的车牌定位算法做了改进,根据彩色图像R、G、B三者的分布特点,采用直接二值化定位车牌。在车牌区域提取之后,针对传统灰度化再二值化进行字符分隔对光线敏感的缺点,本文跳过灰度化过程,基于R颜色分量进行二值化,这种方法很好减少了阳光照射的影响。基于Hough变换,提出了基于双线性插值的两步法的校正方式。本文提出的方法运算过程较简便,代价低,可应用于嵌入式系统低成本硬件上,实现实时车牌识别。