Applied Mechanics and Materials Vols. 411-414

Abstract: We optimized the k-clustering program with MATLAB language in order to improve the stability and quality of k-clustering algorithm in medical CT image segmentation. One hundred and sixty-five head nodule thoracic computed tomography scans are used to test the proposed method and compare with the k-means function of the MATLAB R2012a Statistics Toolbox. We analyzed the difference of the two kinds program running time using single factor variance analysis method and observed the stability and quality of the images segmentation. The experimental results show that the optimized k-means clustering algorithm programming has higher stability and quality of segmentation. In the environment of Windows operation system and hardware of personal computer configuration, the segmenting times are about only one second, significantly lower than the original segmentation procedures. These can eliminate the feeling of waiting and improve the users comfort and efficiency.

Abstract: Genetic Algorithm (GA) is designed to reach the optimal solution via weeding out the worse gene strings based on the fitness function. Thus, more appropriate ones can be found and evaluated through such operation over and over until the optimal solution is researched. GA had been demonstrated the effectiveness in solving the problems of unsupervised image classification, one of the optimization problems of large domain. Two indices DBI and FCMI are analyzed for the cluster centers of spectrum in this paper. To evaluate the consentience between the GA classification and the ground truth in the spectrum distribution of landuse, a SPOT-5 satellite image of Shihmem reservoir is adopted for demonstration.

Abstract: The identifying accuracy of crowd formation recognition models are typically tailored to the type of structure of model, or a specific scene domain. In this paper, we explore the sub-optimal structure of crowd formation model with K2 algorithm. Experimental results show that the recognition model structure taking into account at most two parents performs better than others on a data set generated from Xidan Commercial Street.

Abstract: This paper presented a study on printed character database image analysis based printed document examination in purpose of identifying the printer which created a suspect printed document. It was composed of printed document image acquisition, image pre-processing, feature extraction and classifier. After characters are extracted and recognized in pre-processing, stroke feature sequence of each text block are calculated, and the HU moments of the sequence are also calculated. Finally, the Euclid distance classifier and MQDF classifier are used to recognize the fonts using the above two kind of font feature respectively. Experiments are carried out in a database including 40 LaserJet printers. The experimental results demonstrate the effectiveness of the proposed method.

Abstract: With the rapid development of the remote sensing satellite, the size and resolution of remote sensing image grow increasingly. The evaluation of image quality requires precise information of ground control points extracted from remote sensing image and reference image. Therefore, we propose an adaptive Wallis enhancement based on radiation-parameters to increase the number of ground control points and to improve the matching precision. First, feature vectors of sub-region are constructed by computing image radiation-parameters, and then the sub-region terrain in the remote sensing image can be recognized using nearest neighbor classifier. Second, according to specific type of sub-region terrain, we enhance images using adaptive Wallis filter with local parameters. Finally, two-level matching method is used to extract and match the control points. The experiments show that compared with existing Wallis filter which are based on global parameters, our method gets better results in the detail enhancement on ZY-3 image so that more and higher accurate ground control points can be effectively extracted to achieve the evaluation of geometric precision automatically and accurately.

Abstract: In recent years, with the development of the science and technology, equipment automation has become a trend. IGBT(Insulated Gate Bipolar Transistor) automatic wire system is one of the key equipment, and the vision localization technology as the key technology for improving the performance, has become a key research area. This paper analyzes the principle of wire products, study template matching and geometric shape matching methods based on image processing technology. Experimental studies show that the vision positioning speed and the accuracy of the system has met the needs of wire equipment.

Abstract: According to the license plate recognition problem, this paper did the research about license plate location and characters recognition. It proposed two new algorithms, they separately are license location algorithm based on color segmentation and fault-tolerant characters recognition algorithm based on BP neural network. In the pre-processing stage, it proposed image enhancement algorithm which could make the image more easily analyzed by computer. In the location stage, it made utilization of color and shape information, and then proposed location algorithm. In the recognition stage, it fully made the consideration of characters fault-tolerant, and then made the use of improved BP neural network to recognize characters. Experiments show that the special license plate fault-tolerant characters recognition algorithm is more accurate than the original license plate recognition methods, and its recognition rate has been improved greatly.

Abstract: To distinguish people’s identities, the information is normally included in one gait periodic sequence image. First, the gait energy image for feature extraction of wavelet moments was constructed. After boundary unwrapping, the gait silhouette boundary was extracted and principal component analysis (PCA) was use to obtain its compressed contour features. Then nearest neighbor classifier and support vector machines were applied for classification of these two features. Finally, support vector machine (SVM) on Bayesian rule were used to complete gait recognition with information fusion of different features. The method is evaluated on the National Laboratory of Pattern Recognition (NLPR) gait database and the correct recognition rate is relatively high. The experimental results show that the proposed method has good recognition performance.

Abstract: Palmprint is widely used in personal identification for an accurate and robust recognition. Multispectral palmprint images capture under different illumination, including Red, Green, Blue and Infrared maybe contribute to the recognition results. However, the evaluation of selection and fusion of how this different spectral images can contribute to improve the robustness of the recognition system is imperative. In this paper, a novel wavelet-based multispectral fusion strategy is presented firstly to obtain the fused images; then block singular value decomposition (B-SVD) is applied for feature extraction; Finally back propagation (BP) neural network method is adopted for authentication. The proposed algorithm is evaluated on PolyU database which contains palmprint images from 500 individuals from four independent frequent band. The obtained results show robustness of our multispectral palmprint image fusion and selection model in comparison with the single spectral palmprint image that presented in the literature.

Abstract: Aiming at the problem of slow speed in image matching,anti-interference difference and relatively poor ability to resist deformation,proposed a fast image matching method based on artificial fish swarm algorithm (AFSA). In the same test environment,Compared with the image matching method based on particle swarm optimization (PSO) algorithm and found,the method is superior to image matching method based on particle swarm optimization,in matching speed and noise resistance ability and deformation resistance ability has a marked improvement.