Applied Mechanics and Materials Vols. 278-280

Abstract: Sound recognition based on neural network is a technique that can put a resolution to exceeding artificial identification. Three kinds of neural network recognition models, adopting MFCC and difference MFCC, are discussed. According to six kinds of typical gunshots we design a kind of sound recognition system based on BP neural network optimized by PSO that uses MFCC and difference MFCC as a characteristic quantity to recognize sound signal. In the experiment PSO is used to optimize the network’s initial weights and threshold value. The experiment’s results show that BP neural network optimized by PSO using both MFCC characteristic quantity and difference MFCC characteristic quantity have a relatively lower error and a relatively faster speed than other ways discussed in the article, and the designed system reaches the expected goal.

Abstract: Multi-sensor data fusion provides significant advantages over single source data to achieve an improved accuracy and better precision. Decentralized data fusion approach is one in which features are extracted and processed individually and finally fused to obtain global estimates. The paper presents decentralized data fusion algorithm using factor analysis model. Factor analysis is a statistical method used to study the effect and interdependence of various factors within a system. The proposed algorithm fuses accelerometer and gyroscope data in an inertial measurement unit (IMU). Simulations are carried out on Matlab platform to illustrate the algorithm.

Abstract: It is presented a novel method for image edge detection with information measure and Support Vector Machine, which is called EDWIS (edge detection with information measure and support vector machine). Both the theory analyses and the experimental results show that EDWIS not only can effectively reduce the noises of the image, but also can precisely realize the edge-position, and keep the image edges’ details well. For three kinds of images, two of them including noise, the edge detection results of EDWIS are better than those of Canny, or Sobel differential operator.

Abstract: This paper presents importance of skin texture information in face recognition. To this end, we perform the illumination normalization on face image in order to extract texture information unaffected by illumination variation. And then apply mask image on each illumination normalized face image to obtain the corresponding texture data, which hardly contain the shape information. Face recognition experiments are carried out by using texture data. Experimental results on Yale face database B and CMU PIE database show that the texture information has considerable ability to distinguish subjects in face recognition.

Abstract: Two dimensional barcode has widely applied in identity authentication. In this paper, the image processing methods used in PDF417 barcode recognition were researched, and a quick and effective method to calculate the width of the unit module in PDF417 barcode was proposed. The fast and omni bearing recognition and decoding of the PDF417 barcode was realized.

Abstract: With the ceramics market's developing, the use of image processing and intelligent algorithm is applied to the ancient ceramics recognition and appreciation is one of the most challenging issues in the field of ancient ceramics. Article focuses on selected Ming Qing Dynasty blue and white porcelain as research samples, and explore how to extract the effective image recognition features of ancient ceramics, and to quantify the comparison, given the ancient craft of evaluation index system, and improve the identification of categories of, and appreciation evaluation model to extract special recognition feature, image preprocessing, discussion handwritten the key technology of Chinese character segmentation, feature extraction and classifier design a variety of methods, and non-linear support vector machine analysis method using multiple classifiers based, so that the sample's accuracy greatly improved.

Abstract: We propose a tracking algorithm for a single non-rigid object based on its foreground hue histogram. A tracked region can be described by the foreground hue histogram only calculating foreground object pixels, which can effectively restrain the disturbing of complex background environments. For measuring the object likelihood, we match the foreground hue histogram with that of the tracked object and refer the result of motion detection to encircle the tracked object region as much as possible. During the tracking, we update the hue histogram model for adapting the object appearance variation. The proposed algorithm is realized in the particle filter frame, and the experiments show that it is capable of robustly and accurately tracking a single non-rigid object for the situations of complex background scenes and strong appearance variations.

Abstract: A fast sparse representation face recognition algorithm based on Gabor dictionary and SL0 norm is proposed in this paper. The Gabor filters, which could effectively extract local directional features of the image at multiple scales, are less sensitive to variations of illumination, expression and camouflage. SL0 algorithm, with the advantages of calculation speed，require fewer measurement values by continuously differentiable function approximation L0 norm and reconstructed sparse signal by minimizing the approximate L0 norm. The algorithm obtain the local feature face by extracting the Gabor face feature, reduce the dimensions by principal component analysis, fast sparse classify by the SL0 norm. Under camouflage condition, The algorithm block the Gabor facial feature and improve the speed of formation of the Gabor dictionary. The experimental results on AR face database show that the proposed algorithm can improve recognition speed and recognition rate to some extent and can generalize well to the face recognition, even with a few training image per class.

Abstract: The traditional semi-supervised clustering based on one-class support vector machines used some labeled data called seeds for the clustering initialization. These seeds were partitioned into several initial groups according to their labels and the number of initial groups was equal to the number of clusters. However, the traditional semi-supervised clustering based on one-class support vector machines is sensitive to the initial groups and often obtained the local optimal solutions. In this paper, more initial groups produced by seeds are applied to the traditional semi-supervised clustering based on one-class support vector machines to get more local optimal solutions and the proposed algorithm can combine multiple local optimal solutions to obtain the better clustering performance at last. To investigate the effectiveness of our approach, experiments are done on two real datasets. Experimental results show that the presented method can improve the clustering accuracies compared to the traditional algorithm.

Abstract: This paper proposes a YUV color image super-resolution reconstruction algorithm based on sparse representation. The R, G, B components of color image are highly correlated, three-channel super-resolution independent reconstruction will lead to color distortion, so in this paper the color image is firstly converted to the Y, U, V three channels, and then super-resolution reconstruction. For choosing the regularization parameter, this paper proposes an adaptive regularization parameter method; it has a good inhibitory effect on image noise and adaptive super-resolution reconstruction of color images. The results of experiment show that the proposed algorithm has a better PSNR, compared with bicubic interpolation method and sparse representation. The adaptive super-resolution reconstruction can further improve the quality of the reconstructed image and the method is robust to image noise.