Applied Mechanics and Materials Vols. 543-547

Abstract: In face recognition system, the purpose of gray pretreatment is to denoise and enhance the image. The traditional linear or nonlinear denoising algorithm can bring edge loss, which make it difficulty for the subsequent image segmentation or matching. Although Alpha filter can bring the minimum loss of image edge, the size of filtering window cannot be adaptively changed according to the noise. The Alpha filter is improved on the basis that the information entropy can reflect the noise strength to some degrees. The single pixel entropy in neighborhood is compared with the information entropy average and then the noise infection of neighborhood pixel is determined. Moreover, according to the noise infection, the window size is adaptively adjusted to filter. The results show that the loss of image edge obviously reduces. Because the image size is fixed, we can calculate the integration of normalized image according to cumulative distribution function of the image. Therefore, the image histogram equalization is derived and the image gray is transformed to get the enhanced image. Finally, the results show that the face image after improved gray pretreatment can well ensure the image edge integrity and the face recognition effect is improved by edge feature.

Abstract: In the recognition system of license plate, the detection effect is often influenced by the speed of vehicle, the weather and illumination condition. However, the image edge is less influenced by the above conditions, so it gets more and more attention by using edge detection method to detect license plate. In this paper, three kinds of edge detection method based on partial derivative are compared. Firstly, using the first derivative to get the point set of gray step is discussed and thus the edge is obtained. However, this methods' result is largely influenced by noise. Secondly, adopting denosing theory and second partial derivative to acquire the image edge is represented, but the result shows that this method would filter out some high frequency edges and lead to the edge loss. Finally, the improved algorithm that is the fusion of three aspects: denosing theory, the second partial derivative and linking isolated edge points, is put forward. The result shows that the third algorithm has strong ability to restrain noise. However, at the same time it would smooth some high frequency edges out and lead to the edge loss. However, the third method finally makes isolated points link together, which ensure the integrity of the edge. Therefore, the result obtained by the second partial algorithm is better than the results by the two previous algorithms.

Abstract: With the conflict between storage space of the ship navigation radar system and large raw echoes data and files, a mixed compression algorithm of the bitmap compression + half-byte compression + Double Huffman compression is presented which is based on the characteristics of the text type data. On the basis of the original half-byte compression algorithm, the method combined with the bitmap compression and Huffman improved compression algorithm. If just using the half a byte compression algorithm, the compression ratio can only be 50%; while using the hybrid compression, the total compression ratio can be up to 75%, which save a large amount of storage resources for ship navigation system, and improve the rate of data transmission at the same time.

Abstract: More and more intelligent transportation technologies are applied to license plate detection and recognition that can greatly reduce the burden of traffic management. However, character segmentation of license plate is an indispensable step of license plate recognition. Traditional character segmentation algorithms of license plate mainly use the space between characters of license plate to segment characters, but the license plate cannot be recognized if there are degraded characters or license plate inclination. In this paper, an improved character segmentation algorithm of license plate is proposed. In the improved algorithm, firstly, the noise of precise located license plate is eliminated and then the license plate inclination is tested. When there is license plate inclination, we calculate the inclination angle and use rotation to correct the inclination. So, the problem of license plate inclination is solved in character segmentation. Lastly, the results show that the improved algorithm has better effect than the traditional algorithms and it lays a good foundation for the next research of license plate recognition.

Abstract: Geometric correction is an essential processing procedure in remote sensing image processing. The algorithms used in geometric correction are time intensive and the size of remote sensing images is very large. Meanwhile,the data to be calculated is in huge size and is accumulating rapidly every day. Hence, the fast processing of geometric correction of remote sensing image becomes an urgent research problem. Through the rapid development of GPU, the current GPU has a great advantage in processing speed and memory bandwidth over CPU. It provides a new way for high performance computing. In this paper, we present three optimization solutions based on CPU-GPU hybrid architecture and the analysis of their performances. Experiments are also given and the results are consistent with the analysis.

Abstract: Because of the good parallelism of the generation procedure of digitally reconstructed radiographs, a real-time generation algorithm of digitally reconstructed radiographs based on Compute Unified Device Architecture is presented in this paper. Firstly, the volume data and other input parameters are read and loaded into the GPU; Secondly, the kernel procedure which can be used to simulate the decay process of X-ray in the human body is designed according to the correspondence between X-rays and threads; Finally, the kernel function is executed in parallel by the multi-thread to complete the DRR image generation. The experimental results show that this algorithm uses effectively the parallel computing capabilities of GPU in the premise of ensuring the quality of the DRR, improves significantly the generation speed of DRR, and meets the real-time requirements of digitally reconstructed radiographs in the image-guided radiotherapy.

Abstract: Adaptive algorithm is the key technology in smart antenna array system. Fast convergence and small steady state error will be the main factors in beamforming when applying the algorithm to form optimal weight vector. Based on these facts, various step size and transform domain have been added into the former LMS algorithm. By reusing the array input signals and training sequence as reference signal, convergence can be achieved in a short training sequence, thus the transmission efficiency can be improved. According to MATLAB simulation result, the new algorithm has faster convergence speed and smaller steady state error to achieve an optimal beamforming performance.

Abstract: The paper is devoted to put forward an effective algorithm for removing image impulse noise by improving the traditional median filter. The following approaches are developed in this paper. First, in a small area all the pixels are separated into signal pixels and suspicious noise pixels making use of the extremum and median filtering algorithm .Then suspicious noise pixels are further identified the signal pixels and noise pixels around pixels. If a pixel value is a noise pixel, it will be processed by median filter. A signal pixel will retain the original pixel value. The numerical experiments show that compared with traditional median filter , the details and edge information of the original image are retained when the new algorithm is used , Comparing with extremum median filter algorithm, the ratio of peak signal to noise can increase up to 2.94 at most. As a result, it is an effective algorithm for image denoising that can eliminate impulse noise in the image effectively.

Abstract: It is quite obvious that multi-branch tree search algorithm could be used for higher identification efficiency in RFID, however the branch number choosing depends on the accurate tag number estimation, which is not so easy to get and often cost much overhead. In this paper, we propose an adaptive multi-branch tree anti-collision algorithm. It begins with a simple initial tag number estimation, and then adaptively adjusts the branch groups afterwards. Simulation results indicate that the proposed algorithm can effectively correct the improper branch number and achieve much better performance than the conventional one.

Abstract: AMC technology has been widely applied in the LTE system due to its advantages , but the majority methods of AMC algorithm of CQI reporting ignores the influence of time delay error currently. This article first explores the delay effect of the error of CQI reporting and block error rate of the system through the simulation. Aiming at the influence of time delay error, a CQI reporting mechanism based on neural network is presented in this paper and we found this algorithm effectively solves the problems of the existence of time delay error through simulation on the improved basis.