人脸识别及其在安卓操作系统中的应用英语论文(2).doc_第1页
人脸识别及其在安卓操作系统中的应用英语论文(2).doc_第2页
人脸识别及其在安卓操作系统中的应用英语论文(2).doc_第3页
人脸识别及其在安卓操作系统中的应用英语论文(2).doc_第4页
人脸识别及其在安卓操作系统中的应用英语论文(2).doc_第5页
已阅读5页,还剩1页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

HMM-based face recognition system embedded in the android in the research and its applicationKeywords: HMM face recognition embedded ARM9 0 IntroductionEmbedded face recognition system and compared to the traditional system of identification has a strong advantage, no special acquisition equipment, low cost, simple to use; the same time, face recognition does not interfere with the user, does not infringe the privacy of users, are initiative to identify non-infringement, easy for users to access this main line of embedded ARM9 system development to the theoretical basis of the HMM, shows complete image capture, face detection and recognition functions corresponding to the hardware platform and software modules designed to and implementation process; and image pre-processing is optimized to do floating-point arithmetic, greatly increasing the speed of embedded systems.Part of the system software can be directly applied with the Linux operating system for smart phones, the use of mobile phones and built-in camera, a personal facial feature data can be analyzed and then stored in contrast to the initial face information database, complete the identification function . 1 A system architecture and designThe system uses the Samsung introduced the ARM 920T RISC processor kernel - S3C2410A. Its excellent processing performance should be viewed as the first choice for the development of portable devices. At the same time to meet the smart phones on the video image acquisition needs, the system uses a USB bus-based video acquisition module, and greatly improved compared to the serial transmission of data acquisition rates. The system involves digital image acquisition, processing, storage, transmission and HMM algorithms and other techniques. The system architecture shown in Figure 1. 2 image acquisition hardware designIn view of the traditional high cost CCD image sensor, the relative complexity of the additional circuitry and high power consumption, the system uses OmniVisions OV7640 CMOS chip as the companys image sensor. OV7640 is a low voltage (2.5 V), high sensitivity CMOS image sensor. Real-time acquisition and storage systems need high-speed data transmission, the system hardware with the question put forward higher requirements. The system design, parts acquisition and transmission parts in between with the appropriate buffer. Practice, and supporting the use of OV7640 OV511 extend DRAM chips from the cache role in achieving high-speed digital video images via USB into the ARM processor. OV511 is a dedicated digital camera USB interface IC chip.3 image capture programThe system uses Linux as the operating system platform, operating system migration do not do too much in this introduction.Video4 Linux (short V4L) is a Linux kernel on the video device driver, it is for the video equipment to provide a series of application programming interface unctions, the video f equipment on the market today, including the popular TV card, video capture cards and USB cameras . Video4Linux Linux kernel provides the application interface, program development, the first is based on the Video4Linux API function to design the program.Video4Linux image acquisition based on the program flow shown in Figure 2.4 Image preprocessing and face recognition algorithm and implementation Face recognition process first to determine the input face image or face the existence of the video, if there is further given the position of each face, size and location of each of the major organs of facial information, and based on this information, further extraction of each inherent in the face of personal identity, and has been the face of its library in the face compared to identify the persons identity.Face recognition process can be divided into image pre-processing, face detection and face recognition of three parts. Face detection is the positioning of the matrix to be identified from the face region in the feature area, and separated from each region. Face recognition is based on the existing face database, the output corresponding to the test face in the face database object label. Both as premise and purpose. HMM can be completed as face detection, face recognition can be done, so we will face detection and recognition simultaneously.4.1 Hidden Markov Model (HMM) basic conceptsHMM is a set of statistical properties of the signal characteristics of the model, which consists of two related processes: one is the implicit, invisible to a finite state Markov chain, which has initial state probability distribution function and the state transition probability matrix, the other is a group with state of the probability density function.4.2 HMM model for face recognition A HMM can be denoted by = A, B, , because of its limited character set input V = v1,v2, . vm, so called discrete hidden Markov model.According to the type of state transition, HMM can be divided through the (ergodic) and from left to right (left-right). The former state transition that is arbitrary, can be to own and all other state, which state transition is limited to itself and the next state. Face vertical and horizontal direction from left to right from top to bottom in all regions with the same natural order, can be simulated with the 1D-HMM for face, shown in Figure 3.4.3 facial image feature extractionLet each individual face image width W, height H, is divided into overlapping blocks. The height of the block L, overlapping depth P. Therefore, the face image extracted from the blocks to score the number of observed vector T, and T = (HL) / (LP) +1. Choice of parameters L and P will affect the systems recognition rate, the large value of P increased the depth of overlapping vertical number of feature vectors, the system recognition rate. L, choose more subtle, smaller L can not effectively identify the observation vector; and large L the shear increases the probability of intersecting features. When P is large, the system recognition rate is not sensitive to L changes. Segmentation process shown in Figure 4.4.4 face of the HMM trainingHuman face image database each individual faces a HMM model, with the same person five different face photographs for training. In accordance with sub-block partition method, the resulting 2D-DCT transform coefficients form a vector observation vector sequence. With the observation vector sequence O = o1, o2, ., oT training, obtain the HMM parameters. First HMM model = A, B, is initialized by evenly split from top to bottom to get the training face image data. Model number of states N = 6, with each state of the observation vector sequence has been observed for the initial estimate of the probability matrix B, A and the initial value of at face model from left to right structure is given. Then using the maximum likelihood estimation algorithm (Baum-Welch estimation algorithm) to re-estimate model parameters, detection P (O | ) of the convergence criteria.If (3) conditions, the model has been convergence, the end of the training iteration process; otherwise continue to the next training.Here, C is the pre-specified threshold.4.5 Face Image RecognitionFace image to be identified for the training process to extract the same observation vector sequence, the probability of observing the vector sequence by the HMM model face image calculated, namely: When satisfies (4), the recognized face image database in the corresponding face of the k individuals face are identified.Experiments show that this algorithm is easy to implement real-time processing, free from the impact of changes in facial expression, anti-noise ability, good robustness. However, the illumination problem in face recognition and attitude problem needs further study.5 Conclusions HMM algorithm based on ARM9 embedded face recognition system with a small, small amount of calculation, fast speed, stable performance, etc., to meet peoples demand for recognition device miniaturization. I believe in the near future, the face recognition system based on embedded in the security, authentication, access control systems, intelligent attendance so widely used. Google released Android 4.0 system supports face recognitionKeywords: Face Recognition Google Apps Jobs System equipped with Android 4.0 smartphone Samsung GALAXY Nexus October 19 morning news, Google and Samsung at 10 am this morning held a news conference in Hong Kong, the official release, code-named ice cream sandwich generation of Google Android 4.0 system and smart phone Samsung GALAXY Nexus. Sina science and technology throughout the video broadcast of this conference. Google and Samsung had planned on October 11 in San Diego (San Diego) at the U.S. wireless show (CTIA) and posted on the Android 4.0 system GALAXY Nexus smart phone, but due to former Apple CEO Steve Jobs (Steve Jobs ) October 6, sudden death, mourning Jobs in the industry, so Google and Samsung postpone the conference. Google has a new generation of Android 4.0 system designed UI, interface to simplify, speed capacity building. The new system is built-in wallpaper, use the phone button on the touch will be glow, and increased voice input, face recognition, photo editing, traffic monitoring, Android Beam function. The new system allows users to applications on the phone and plug-ins on the main screen, the user need not enter into the application inside the thumbnail you can see some of the latest information. New version of the system to achieve a user interface sliding around, looking for user-friendly application. In addition, users can also expand the application interface for free. Latest system, the folder can drag the program to change the program location; while folders can also be fixed on the taskbar, user-friendly call at any home screen. The new system also convenient for users to switch between any application, by moving to the left, users can open and close the application. Android 4.0 system has a screenshot function, screen shots, after the picture will automatically enter the album. Meanwhile, the Center notified the user to enter the Andriod faster, so users can drag some of the notice to remove the BTS. In the SMS function, Android 4.0 features an improved keyboard, not only the response speed faster, more accurate error correction, copy and paste tool has also been improved. In addition to the keyboard, but also send text messages to support voice input. Android 4.0 is another highlight, an increase of facial recognition to unlo

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论