【作者】 魏巍；

【导师】 刘富；

【作者基本信息】 吉林大学 ， 模式识别与智能系统， 2012， 硕士

【摘要】 在当今网络信息技术快速发展的时代，个人身份信息泄露事件频频发生，传统身份识别方式的可靠性受到了高科技犯罪的挑战，因此，以人体所特有的、不易窃取的生理特征进行身份鉴别的生物识别技术引起了国内外相关研究机构的广泛关注。手形识别作为一种常见的生物识别技术，主要以手形轮廓的几何尺寸为特征，具有算法简单、识别速度快，易于和掌纹、静脉等特征融合，实现多模态身份识别等优点。虽然手形识别技术的理论知识已经日趋完善，相应的识别产品也在一些公共场合发挥着重要的作用，但是，这些技术和产品都以PC机为依托，识别装置体积大、成本高，不利于商业化发展。根据当前生物识别技术的发展趋势，利用ARM、DSP等专用图像处理芯片建立的，集手形图像采集、处理、识别于一体的嵌入式手形识别系统具有广泛的研究价值和广阔的商业前景，是未来手形识别产品的发展方向。要实现一个基于DSP的手形识别系统，首先就要解决手形图像采集以及手形图像处理算法在DSP上的实现，本文正是围绕这个两个问题，进行了以下几个方面的研究：（1）手形图像采集方案的设计。系统最初选择了数字摄像头OV7670获取图像，5509A控制完成手形图像采集和处理，但是在处理速度、存储能力上达不到性能要求。在参考了TI提供的相关技术文档后，选择了DM642+TVP5150A的解决方案，TVP5150A从CCD提供的模拟信号中解码出手形图像数据，DM642完成数据的读取并进行后续的处理。（2）基于DSP/BIOS的手形图像采集驱动开发。DSP/BIOS是TI公司专门为其生产的DSP芯片提供的一个嵌入式实时操作系统内核，方便用户编写多任务程序，在大量阅读了DSP/BIOS相关知识的基础上，充分利用TI提供的芯片驱动开发包（DDK）中的类/微驱动模型和API函数设计出了相应的采集程序，完成手形图像采集。（3）手形图像处理算法的DSP实现。在DSP系统中，数据的格式、存储方式都有所不同，处理速度也比PC机慢，因此必须对以前的算法做一些调整和修改才能在DSP系统上实时运行。在手形图像的增强处理中，本文采用了一种3×3的快速中值滤波算法；阈值分割、轮廓提取都采用了常规的、复杂度低的方法；在角点检测时提出了一种基于相对距离的检测方法、该方法计算量小，准确度高。整个手形图像处理过程很好的兼顾了处理的效果和实时性。最后，结合实验结果从实时性、CPU处理能力等方面对系统性能进行了详细的分析，说明了本文所采取的手形图像采集、处理系统的可靠性。更多还原

【Abstract】 Nowadays, with the rapid development of network IT, the matter of personalidentification information being revealed occurs frequently, as a result, the reliabilityof traditional identification methods are challenged by high-tech crime.Therefore, thebiometric technology utilizing physiological features of human beings, which areunique to human body and not easy to be stolen, has been accepted widely attentionby domestic and foreign related research institutions. Hand-shape recognition, as ausual kind of biometric technology, which is mainly characterized by the geometrydimensions of hand-shape contour, has the advantages of simple algorithm, highrecognition speed and easy to be fused with features of palmprint and vein to achievemulti-modal identification.The theoretical knowledge of hand-shape recognition has been maturing, and thecorresponding identification products also play an significant role in several publicplaces.However,these technology and products are all based on PC,which makes theidentification device has huge volume, high costs and not beneficial to thedevelopment of commercialization. According to the current development trend of thebiometric technology, the embedded hand-shape recognition system, which sets thehand-shape image acquisition, processing and recognition in a group utilizingdedicated image processing chip, such as ARM and DSP, has widespread researchvalue and broad commercial prospects and is the development orientation ofhand-shape recognition products in future.To achieve a hand-shape recognition system based on DSP, firstly, we must try tomake the acquisition and processing algorithm of hand-shape image beingimplemented on DSP.This article has done the following aspects of research aroundthe two issues.（1） Design of hand-shape image acquisition program.We initially choose the digital camera OV7670to acquire hand-shape image and5509A to control the hand-shape image acquisition and processing.However, thischoice can’t satisfy the performance requirements of processing speed and storagecapacity. After consult related technical documentation offered by TI Company, we finally choose DM642and TVP5150A to solve this problem. In this program, TVP5150Ais used to decode the data of hand-shape image from the analog signal offered by CCDand DM642is used to read the data and progress the following treatments.（2） Exploitation of the hand-shape image acquisition driven system.DSP/BIOS, an embedded real-time operating system kernel, which is speciallydesigned by TI Company for its DSP chip, are convenient for users to write multi-taskprograms. On the basis of reading a large amount of knowledge of the DSP/BIOS, wetry our best to make full use of class/mini driver model and API function from the chipdriver development kit (DDK) offered by TI company to design the hand-shape imageacquisition program, by utilizing which we complete the hand-shape imageacquisition.（3） Achievement of hand-shape image processing on DSP.The format of data and the storage method are different in DSP chip, and itsprocessing speed is slower than PC.Therefore, we must do some adjustments andmodifications for the previous algorithm to make sure it can be operated on DSPsystem. In hand-shape image enhancement, this paper uses a3*3fast median filteringalgorithm and uses conventional and lower complex method for thresholdsegmentation and contour extraction. This paper proposes a relative distance-baseddetection method in corner detection stage, which has fewer amounts of computationand higher accuracy. Not only the processing effect but also the real-time quality hasbeen taken into account perfectly in the whole process of the hand-shape imageprocessing.Finally, according to the experimental results, we have made a detailed analysisfrom real-time quality and the CPU processing capacity to illustrate the reliability ofthe hand-shape image acquisition and processing system in this paper.更多还原