《人脸识别及其在安卓操作系统中的应用英语论文(2).doc》由会员分享，可在线阅读，更多相关《人脸识别及其在安卓操作系统中的应用英语论文(2).doc（6页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、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,
2、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 compl
3、ete 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 s
4、oftware 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 archi
5、tecture 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, t
6、he 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 Figu
7、re 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
8、 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
9、 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 n
10、ot 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 an
11、d 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 an
12、d 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 in
13、herent 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 t
14、o 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 re
15、cognition 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 Mar
16、kov 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
17、,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
18、 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
19、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
20、 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
21、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
22、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 o
23、bservation 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,
24、 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
25、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 algor
26、ithm 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 recognitio
27、n 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 at
28、tendance 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
29、 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 sho
30、w (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
31、 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
32、 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 fo
33、r 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 th
34、e 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 i
35、mproved 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 unlock features. T
36、hrough the camera to recognize human faces, the program will determine whether to unlock the phone, if not the owner himself, the system will respond to prompts. In addition, the new system, the browser, photo albums and other features also made large improvements. The new system browser supports 16
37、 windows open simultaneously, to achieve true multi-tasking. Users browsing the same time, can easily be stored bookmarks. The new system also adds a new flow of new early warning and monitoring capabilities to help users check a certain period of time in traffic usage of each application. Camera fu
38、nctions, the lock-screen mode, the user can move to the left turn directly into the camera button state, the pictures can be promptly posted to social networking, email, etc. If someone in the face the camera, the system will automatically recognize. In addition, the camera also supports the panoramic camera, and images can be edited. The new system also supports Android Beam functions, based on NFC near field communication technology, two back to back touches the mobile phone can easily exchange data, such as web pages, maps, games and so on. (