多重频谱滤波法改进体全息存储图像相关识别  

Multiple-spectrum Filtering for Improvement of Volume Holographic Correlator

作　　者：王大勇[1] 陈孙征[1] 袁玮[1] 万玉红[1] 江竹青[1] 陶世荃 

机构地区：[1]北京工业大学应用数理学院,北京100022

出　　处：《光电子．激光》2004年第6期719-723,共5页Journal of Optoelectronics·Laser

基　　金：国家重点基础研究与发展计划(973)资助项目(G19990330);北京市教委科技发展计划资助项目(2002KJ002);北京市科技新星资助项目(952872100)

摘　　要：采用体全息存储器建立图像数据库,存储器件是立方形掺铁铌酸锂(Fe:LiNbO3)晶体,1020幅汉字图像通过水平和垂直的角度复用技术以及顺序曝光法记录在晶体的同一点中,可以实现并行相关运算。提出了多重频谱滤波方法,有效地抑制直流分量和噪声,使相关峰陡峭,自相关峰值与背景平均值之比约为4.2,极大地提高了体全息存储相关器的识别精度。An image database was constructed by using the volume holographic memory. The memory device was a cubic Fe: LiNbO3 crystal. 1020 Chinese images were stored into a single location in the crystal through the angularly-multiplexing and subsequence exposure. An input object could be correlated with all the stored images simultaneously to achieve pattern recognition. Also, a novel multiple-spectrum filtering was proposed to suppress the DC term and the noise. The experimental results show that the method makes the auto-correlation peak sharper, the ratio of the autocorrelation peak over the mean of the output background is about 4.2, and the discrimination of the system is improved.

关 键 词：全息存储器 图像数据库 角度复用 顺序曝光法 并行相关运算 多重频谱滤波 自相关峰值 识别精度 

分 类 号：TP333.42[自动化与计算机技术—计算机系统结构]