【作者】 滕东东；

【导师】 王彪；

【作者基本信息】 哈尔滨工业大学 ， 工程力学， 2007， 博士

【摘要】 信息技术和计算机技术的高速发展对信息存储技术提出了更高的要求。体全息存储技术以其存储密度高、存储容量大、数据传输速率高、数据搜索时间短等优势成为一种颇具潜力的海量信息存储技术。而基于体全息的相关识别技术是一种新兴的光学模式识别技术。它通过角度复用光学体全息存储技术将图像库存储在光折变晶体中,建立识别模板库。当输入载有待识别图像信息的物光时,产生一系列与模板库对应的相关峰,各个相关峰强度正比于待识别图像与相应模板的相似度。传统的目标识别方式受限于数字计算机的串行依次处理,其识别时间与模板库的大小成正比,尤其是对海量模板的匹配,其时间开销是巨大的。而体全息相关识别技术是一种并行处理技术,待识别目标只需经过一次并行处理就能完成与模板库中所有模板的匹配处理,其识别时间与模板库大小无关。本论文主要以掺铁铌酸锂晶体为存储介质,进行基于光折变体全息的相关识别系统的研究。在传统热固定技术的基础上,综合考虑动态范围和热固定效率对存储过程的影响,对各种参数进行了优化处理。当存储介质同一体积内存储的全息图数量很大时,在采用顺序曝光时序进行存储的过程中,最早存储全息图的曝光时间较长,光致散射比较严重,不利于识别准确率的提高。为了解决这个问题,并得到更理想的热固定效率,本论文还介绍了分批热固定理论,研究了批间光擦除时间常数对存储过程的影响,最终实现了分批热固定后衍射效率的均匀化。以薄晶体为存储介质,根据离焦相关识别系统的位移不变性,论文进行了以人耳为识别目标的实时图像识别。在存在一定程度平移、变形或旋转等畸变的情况下,准确实现了同一类别图像不同模板的归类识别。对于非实时图像,我们以厚晶体为存储介质,利用其大的动态范围,大大提高了同一体积内可以存储的识别模板的数量。在不考虑图像畸变的情况下,论文第四章采用相关系数模板的识别算法,实现了1000幅图像分批热固定后的并行识别。为了提高识别系统的实用化,我们在第五章进行了可携式识别系统的研究。采用反射式光路,用会聚的参考光进行存储,将参考光光路去除后,保留激光光源和物光光路进行单独的设计和包装,得到尺寸为350×200×70mm3的可携式相关识别系统。我们还在第五章对相关识别系统进行了一些改进。首先,提出了分维角度-空间复用的存储方法。在相关识别系统中,这种复用方法可以有效的提高存储密度,更充分地利用晶体的动态范围。其次,为进一步提高体全息相关识别系统的识别速率,我们以端面为正方形的长条状Zn:Fe:Mn:LiNbO3晶体为存储介质,在物光固定不动的情况下实现了多个空间位置处的分维角度复用存储。在识别时,输入图像可以与晶体内多个位置处存储的7500幅图像同时进行相关识别,得到了更高的识别速率。更多还原

【Abstract】 With the developing of information and computer technology, the new storage system is needed urgently. Volume storage has become a potential storage means for the merits of high package density, high storage capacity, fast data gransfer rates and short access time. Accordingly, the correlator based on volume hologram is a rising optical pattern recognition technology. Fistly, images for recognition need to be storaged into the refractive crystal with angular multiplexing. All these images serve as template during the later recognition process. Then with the incidence of object light loading the image to be recognized, a sequence of correlation peaks is diffracted, which respond to the image stored. The strength of the correlation peak is proportion to the similarity of image input and that stored. Relative to the conventional target recognition by couputer, the processing progress is parrelelling and time needed has nothing with the number of templates.In our dissertation, hologams are stored in doped LiNbO3. To solve the erasure problem of optical grating stored during the correlation process, we adopt the technology of thermal fixing. Combining the influence of dynamic range and the eficency of thermal fixing, we optimized all the relative parameters and obtain better diffraction efficiency. When the number of images stored as templates is very large, the hologram firstly stored needs a rather long exposure time during the sequential exposure recording. This leads serious light-induced scattering which does great harm to recognition. We introduce batch thermal fixing and study the infulence of inter-batch optical erasure time constant.For real-time recognition with ears of human being as the targets, we take thin crystal as the recording medium. According to the shfiting invariance of defocus volume holographic correlator, we realize the exact subsumption of the inputting real-time image. For the nonreal-time image, we take thick cryastal as the recording medium. Due to the large dymanic range, the number of template stored can be rather large. Without the considering of abberration of image, taking correlation coefficient match plate as the recognition algorithm, 1000 templates were recognized accurately and rapidly in the fourth chapter. For the practicality of volume holographic correlator, we complete a portable correlator system in Chapter 5. In this system, the reference light is convergent beam and reflecitve geometry is adopted. Removing reference light path, we package the laser and information light path and the dimension is about 350×200×70mm3.Furtherly, fractal/space multiplexing is put forward in the fifth chapter. To the voluem holographic correlator, it could improve the storage density and make use of the dynamic range more efficiencly. At last, taking Zn:Fe:Mn:LiNbO3 bar as the recording medium, we store hologarm in a few space with angular multiplexing withour information light path fixed. Thus, with the incidence of information light to be recognized, the parallel correlation with 7500 templates sotred in all the space is realized.更多还原