Applied Mechanics and Materials Vol. 441

Abstract: Boundary localization is one of the key issues for reliable iris recognition system. For non-ideal iris images, eyelashes or eyelids occlusions and low contrast between iris and sclera will lead to inaccurate boundary localization. Specifically, if the intensive transition from iris to sclera is too smooth, outer boundary localization will be very difficult. To stress the problem, in this paper the boundary localization method is proposed in which nonlinear gray transformation is innovated in outer boundary localization process. The experimental results depict that our algorithm have improved the localization accuracy for non-ideal iris compared to the classical algorithms.

Abstract: The extraction of major colors is the basis of image processing. In the process of camouflage pattern painting design, whether major colors are extracted rapidly and precisely are very important to the application. Histogram clustering and K-means clustering and ISODATA clustering are three widely using extraction algorithms. To test and compare the accuracy of the three algorithms, the definition of average color difference is introduced. Then, two representative forest land background images are used to compare the performance of three algorithms. The result shows that, ISODATA clustering algorithm is superior to other two kind of clustering algorithms.

Abstract: With the rapid development of information technology, data grows explosionly, how to deal with the large scale data become more and more important. Based on the characteristics of RDF data, we propose to compress RDF data. We construct an index structure called PAR-Tree Index, then base on the MapReduce parallel computing framework and the PAR-Tree Index to execute the query. Experimental results show that the algorithm can improve the efficiency of large data query.

Abstract: In order to improve the measuring precision of the one-dimension line-matrix CCD, a new sub-pixel edge detection method is proposed. Firstly the date collected by the line-matrix CCD is converted to digital data by virtual oscillograph. Base on threshold comparison, the fitting part of the edge signal is picked up for further processing; secondly, the pixels are expanded following the edge direction of the edge point, and the edge signal is fitted through the two order multinomial; finally, comparing the two order curve with the threshold voltage through the least square fit method, the precise position of the edge point can achieve the sub-pixel edge location precision.

Abstract: This paper aims to present a review of recent techniques in high dynamic range imaging (HDRI), which was the topic in research areas including image processing, computer graphics, and photography. HDRI or just HDR is a set of techniques that allows a greater dynamic range between the lightest and darkest areas of an image than current standard digital imaging techniques or photographic methods. HDR imaging technologies will spread its sphere of influence in imaging industry for its high quality and its powerful expression ability, including digital cinema, digital photography and next generation broadcast. This paper discusses recent advances, future direction in HDRI. The major goal of the paper is to provide a reference source for the researchers involved in HDRI, regardless of particular application areas.

Abstract: Over the last ten years, considerable progress has been made on the new hand-based biometric recognition, such as palmprint and hand vein. During this period, it has been proved that Finger-Knuckle-Print (FKP) can be used as a biometric identifier. In this paper, we present an effective FKP identification method based on Local Binary Pattern (LBP), whose idea is to divide the region of interest (ROI) of FKP into a set of sub-image blocks, which can be applied to extract the local features of the FKP. After that, LBP histograms of image blocks in a FKP ROI image are connected together to build the feature vector of the FKP ROI image. In the match stage, histogram intersection distance is applied as the similarity measurement between sample and template. Experimental results conducted on a database of 165 persons (4 fingers per person) show that the proposed method is effective.

Abstract: We propose a practical image retrieval scheme to retrieve images efficiently. We propose a scheme using color and texture features and address the unique algorithm to extract the color pixel features by the HSV color space and Tamura features of the texture features. The proposed scheme transfers each image to a quantized color code using the regulations of the properties in compliance with HSV color space model and then employing the quantized color code along with Tamura features of texture features to compare the images of database. Experimental of the proposed scheme on demonstrate more efficient and effective than the conventional schemes.

Abstract: In order to study the approximation performance of general regression neural networks, the structure and algorithm of general regression neural networks are first introduced. Then general regression neural networks and back propagation neural networks improved by Levenberg-Marquardt algorithm are established through programming using MATLAB language. A certain nonlinear function is taken as an example to be approximated by the two kinds of neural networks. The simulation results indicate that compared with back propagation neural networks, general regression neural networks has better approximation precision and faster convergence speed, which means it has much better approximation ability than back propagation neural networks. Therefore, for more complex function approximation, general regression neural networks is recommended. It can reduce the complexity of neural networks and it is also easier to design.

Abstract: According to the high speed of data arriving, a large amount of data and concept drifting in the stream model, combining the techniques of rough set theory, neural network and voting rule, we put forward a new data stream classification model, which is a multi-classifier integration based on rough set theory, neural network. Firstly, it reduces all attributes using rough set theory; secondly, it constructs base classifiers on the data chunks after the reduction of attributes using the improved BP neural network; finally, it fuses various base classifiers into an ensemble by voting rule. Through applying the model to classify data stream, the experiment results show that the ensemble method is feasible and effective.

Abstract: The full text retrieval system can receive constant feedback in the form of user behavior. In the case of a search engine, each user will immediately provide information about how much he likes the results for a given search by clicking on one result and choosing not to click on the others. This paper will look at a way to record when a user clicks on a result after a query, and design a Click-Tracking Network. Then training it with BP neural networks to intelligently improve the rankings of the results for users. Finally, we implement a search and ranking system content-based ranking and improve the search and ranking with neural network. By experiments we have shown good results.