A fast eye opening and closing state recognition methed is presented to solve the problem of driver fatigue detection. Firstly, we identify the face region through the skin model. After pretreatment, we carry out Gabor transform. By choosing appropriate scale and orientation parameters, the region of eye has obvious characteristics of the gray. The region of eye has obvious peak after horizontal integral projection of the transformed image, we can recognize eye opening and closing state through the peak-to-average ratio. This way breaks through the limitation that the traditional integral projection method can only location the eye. Experiments show that the accurate of this algorithm is higher, and has good robustness to the changes of illumination.

Because of large size and high price, the commercially available lockin amplifier or virtue lock-in amplifier cannot be integrated into portable devices. By using signal correlation detection principle, a portable quadrature lock-in amplifier is designed and developed based on ARM7 processor, which consists of two PSD(Phase Sensitive Detectors) with quadrature reference signals. Hardware modules are welded and debugged, including reference signal generator, analogue multiplicator, lowpass filter, etc. In order to prove the function of each module, output waveforms are measured and taken. Using the output sine-wave signal from a commercially available signal generator as the signal to be detected, the relations between input and output of the two PSDs are calibrated. The characteristics involving measurement scale, mean error, frequency selective property and detection limit are intensively tested and characterized. Experimental results show that, the detection limit on signal amplitude is less than 5 mV, the measuring scale is 5~3 000 mV, the absolute mean error is less than 10 mV, the mean error relative to the measuring scale is smaller than 1%, the allowed bandwidth of the detected signal is 0~50 kHz, and the 3 dB bandwidth is less than 1 Hz. Because of good frequency selectivity and high accuracy, the proposed portable lockin amplifier exhibits potential applications in weak signal detection, industrial control and instrumental devices.

To solve the slowness and low integration problems of existing algorithms, based on existing technical researches, an improved algorithm is proposed. First, segment the foreground with image segmentation algorithm Grabcut. Second, generate trimap as the input of matting algorithm Shared Matting on the basis of segmentation result. Third, obtain the final matting result. For characteristics of the system, Grabcut algorithm is optimized with downsampling technology to accelerate the segmentation speed and concurrently accelerates trimap generating algorithm and Shared Matting to further improve the processing speed. It is proved by experiment that the system has a strong applicability and can produce an ideal matting effect with less interaction in a short time.

Previous matrix factorization algorithms can not guarantee the nonnegativity of results. Inspired by the nonnegative character of NMF (Non-negative Matrix Factorization), NMF is utilized to detect shadow areas. The BNMF (Block Nonnegative Matrix Factorization) is introduced for shadow detection. The brightness features of shadow points of training samples are extracted by NMF or BNMF method. The shadow areas of testing sample are recognized by features. The results demonstrate that the approach preserves edges clearer than the method based on singular value decomposition.

To improve the accuracy of 3D face reconstruction, a novel coarse-to-fine 3D reconstruction method is proposed. With salient points in each region, a rough shape is reconstructed based on SRSD (Surface Reconstruction from Sparse Data) method. Since the shape reconstructed is a rough approximation of real model, the following reconstruction based on this shape is more accurate. Taking the rough shape as reference model, distance of feature points between 2D face and the projection of reference model can be calculated. Then deforming factors are computed and the final precise shape is reconstructed. Since the roles of each reconstruction step are different, parameters in each step are selected in different ways. Experimental results show that this method can effectively improve the accuracy while having similar computation speed.

We proposed the idea of modeling the large value payment system of China by means of the theory and methods of complex networks to solve the problem of measurement of the node strength distribution of this network. Based on the regional payment data, through this study we found that two network nodes strength distribution follows the power law distribution, but the exponents are different. These two exponents will vary with the setting threshold value. They all do not have small world characteristics.

In order to better control the risk of payment system, based on the existing measure method analysis of the liquidity requirements on the daytime in payment system, relaxing the assumption of no other risk and delayed payment, using the method of dynamic game, the results are found to determine whether a bank should adopt delay payment basis, we put forward a measure of the size of the payment system liquidity risk indicators. Certain theory reference is provided for the central bank when it formulates relevant daytime credit policy to balance liquielity risk and the efficiency of payment system.

To solve the bottleneck problems of the traditional relationship database in big data storage and access efficiency, this paper presents an efficient approach to store big earth data based on NoSQL (Not only SQL) database; MongoDB, a typical software of NoSQL, is compared with RDBMS (Relational Database Management System） in data storage and query performance. The result shows that NoSQL databases have the significant performance advantages in data adding, querying compared with RDBMS. Especially for the mass unstructured and semistructured data, the performance advantage is more obvious.

To solve the problem that traditional FAT format file data access method is unable to adapt to the query operations of the large\|scale catalog file in large\|scale mass storage device as the traditional data access method of FAT format file is sequential access, the wear-leveling algorithm is proposed, optimizing wear block of flash in NandFlash storage medium using bitmap storage technology is realized, and memory index and idle directory of an embedded file system are optimized. A simulation program is developed, and simulation experiments are carried out based on ARM9 MagicARM2410 board. The result shows that the basic functions of the file system can be realized, and the feasibility is verified.

In order to quickly determine the Intervisibility and visible range of height field scene, to meet the requirements of interactive applications, a novel method is proposed for low complexity height field with a simplified convex hull and a binary tree. Through a linear scan in eight directions on height field determined each maximum azimuth, a convex hull tree structure is constructed Information on the convex hull tree is analysed and simplified, Binary tree structure is treated as a storage structure for direction visibility scan, which can effectively improve search efficiency in the lighting calculation for visible point determination. Combined with indirect lighting calculations based on shadow map, the interactive frame rate of low complexity dynamic height field rendering is achieved, reducing the computational cost effectively.

In order to improve the efficiency of building concept lattice in Chinese domain ontology learning based on formal concept analysis, we applied the concept lattice reduction theory to the process of building concept lattice. The main idea is that we reduce the objects and attributes of the obtained formal context which is based on semantic dependency analysis, adopte the Godin algorithm to construct concept lattice based on formal context reduced, and repaire the concept lattice with the reparation theories. The article takes the asclepiadaceae plants ontology construction as an example to verify this method. The experiment results show that the efficiency of concept lattice construction increase by 70%. It improves the efficiency of ontology construction.

When traditional FCM（Fuzzy C-Means) clustering algorithm applied to image segmentation, because of not combining with image spatial information, it is sensitive to noise, can not produce ideal image segmentation. An improved FCM algorithm based on neighboring membership constraint for image segmentation is proposed, which improves the robustness of image noise through adding the neighborhood spatial information to FCM objective function to constrain the membership function. The results of segmentation are expected. Experimental results indicate that this algorithm has a strong ability to inhibit noise, and can get better segmentation effectiveness.

In order to overcome the shortcoming of traditional CAI（Computer Assisted Instruction）, the main module in double active ICAI（Intelligent Computer Assisted Instruction）system was constructed. A student model in this ICAI system is, all of parameter lists in student model are described. On the designing of diagnostic module and the process of data acquisition was emphasized by diagnosticing. The cognitive student model is the foundition of ICAI system. The ICAI can stimulate interest and motivation, and can improve the effects of instruction.

It is better for teachers and students in education to provide the corresponding teaching resources, share resources of teaching, enable students to better grasp the relevant knowledge. Based on ASP (Active Server Page) technology, we put forward a kind of combination of SQL (Structured Query Language) Server 2005 database, design a network teaching platform through the multimedia technology and application of network teaching platform to realize its dynamic characteristics. Practice proves that a better communication between teachers and students can make the teaching obviously efficient.

In order to improve the accuracy of collaborative filtering, the paper proposes a new collaborative filtering based on the dual-threshold neighbors finding method in the perspective of how to find the truly relevant user/item group. This method can take full advantage of existing sparse user-rate matrix to find some users or items which are strong relative to the active user/item, and they can participate in the progress of calculating predicate rate. The experimental results show that the recommendation accuracy of the new algorithm is better than traditional collaborative filtering and some improved algorithms.

It is difficult to control the quality problem of a military system design, in order to solve this problem, based on reference CMM (Capability Maturity Model), a model of military software development capability maturity mode is presented. The theory can provide technical support for establishment of design project management standard of each stage for military software development, to provide theoretical support for the military software project management, military software development risk reduction, scientific control software development phases of fiscal expenditure, the actual development can be applied to military software development project.

Because of the contradiction between one way and trapdoor when we construct public key cryptography will turn up. a new type of public key cryptography with oneway shell core function is proposed. According to the designing ideas and structural characteristics of HFE public key cryptography, the paper presents a scheme of one-way shell core function and shows that the function can resist the present main attack methods. Comparing with the traditional public key cryptography, oneway shell core function has the characteristics of wider inclusivity, more change and higher security. The function provides people with a public key cryptography of more flexibility.