As the "system-on-a-chip" concept is rapidly becoming a reality, time-to-market and product complexity push the reuse of complex macromodules. Circuits combining a variety of the macromodules (micro-processo...
详细信息
ISBN:
(纸本)9780818683596
As the "system-on-a-chip" concept is rapidly becoming a reality, time-to-market and product complexity push the reuse of complex macromodules. Circuits combining a variety of the macromodules (micro-processors, DSPs, programmable logic and embedded memories) are being reported by a number of companies. Most of these systems target the embedded market where speed, area, and power requirements are paramount, and a balance between hardware and software implementation is needed. Reconfigurable computing devices have recently emerged as one of the major alternative implementation approaches, addressing most of the requirements outlined above.
The performance of K-means clustering algorithm is poor for high dimensions data set. The goal of this paper is to reduce the high dimensional data to a meaningful low dimensional data representation, so that the effi...
详细信息
The performance of K-means clustering algorithm is poor for high dimensions data set. The goal of this paper is to reduce the high dimensional data to a meaningful low dimensional data representation, so that the efficiency of clustering algorithm will be elevated. Hence to improve the efficiency of clustering analysis, unsupervised quick reduct algorithm (USQR) is used for selecting the features from high dimensional data. Then the selected features are used to find the initial centroid using k-MAM initialization technique for k-means. The initial centroids are finally used to find the clusters. The results are compared to k-means and k-MAM with USQR so that outperforms well, in terms of accuracy and number of iterations compared to the k-means, for high dimensional data.
This paper presents an alternative algorithm for integrating the existing knowledge of a supervised learning neural network with the new training data. The algorithm allows the existing knowledge to age out in slow ra...
详细信息
This paper presents an alternative algorithm for integrating the existing knowledge of a supervised learning neural network with the new training data. The algorithm allows the existing knowledge to age out in slow rate as a neural network is gradually retrained with consecutive sets of new samples, resembling the change of application locality under a consistent environment. The algorithm also utilizes the contour preserving classification algorithm to increase the accuracy of classification. The experiment is performed on 2-dimension partition problem and the result convincingly confirms the effectiveness of the algorithm.
A stereo obstacle detection method based on images symmetrical move and cell partition is presented in this paper. This algorithm gets each cell's optimum offset by calculating the minimum non-similarity of cell, ...
详细信息
ISBN:
(纸本)9781424435036
A stereo obstacle detection method based on images symmetrical move and cell partition is presented in this paper. This algorithm gets each cell's optimum offset by calculating the minimum non-similarity of cell, and merges Cell of Interests (COIs) to calculate obstacles in the most probably obstacle regions, and gives information of these obstacles. The paper introduces the perspective projection model at first, and then gives two conclusions which are the basic of the algorithm, the part of cell partition and cell optimum offset calculation are presented. Finally, algorithm performance tested by experiment on the test vehicle JJUV-I is shown in the paper.
This paper proposes a two-stage global motion estimation method using a novel quadtree block-based motion estimation technique and an active mesh model. In the first stage, motion parameters are estimated by fitting b...
详细信息
ISBN:
(纸本)0769511104
This paper proposes a two-stage global motion estimation method using a novel quadtree block-based motion estimation technique and an active mesh model. In the first stage, motion parameters are estimated by fitting block-based motion vectors computed using a new efficient quadtree technique, that divides a frame into equilateral triangle blocks using the quadtree structure. Arbitrary partition shapes are achieved by allowing 4-to-1, 3-to-1 and 2-1 merge/combine of sibling blocks having the same motion vector. In the second stage, the mesh is constructed using an adaptive triangulation procedure that places more triangles over areas with high motion content, these areas are estimated during the first stage. Finally the motion compensation is achieved by using a novel algorithm that is carried by both the encoder and the decoder to determine the optimal triangulation of the resultant partitions followed by affine mapping at the encoder. Computer simulation results show that the proposed method gives better performance that the conventional ones in terms of the peak signal-to-noise ratio (PSNR) and the compression ratio (CR).
Our prior work [1] presented a decentralized algorithm for coordinating the construction of a truss structure out of multiple components. In this paper, we discuss adaptation in decentralized construction. We partitio...
详细信息
Our prior work [1] presented a decentralized algorithm for coordinating the construction of a truss structure out of multiple components. In this paper, we discuss adaptation in decentralized construction. We partition construction in two tasks, tool delivery and assembly. Each task is performed by a networked team of specialized robots. We analyze the performance of the algorithms using the balls into bins problem, and show their adaptation to failure of robots, dynamic constraints, multiple types of elements and reconfiguration. The algorithms can be used for general types of source elements.
To recognize an individual's membership in a face group, but without revealing the individual's identity and without restricting the members of group, this paper introduces an SVM classification tree algorithm...
详细信息
ISBN:
(纸本)0780383591
To recognize an individual's membership in a face group, but without revealing the individual's identity and without restricting the members of group, this paper introduces an SVM classification tree algorithm conducting face membership authentication in terms of the "divide and conquer" strategy, and recognizing the membership in a face group whenever the member or the size of group changes dynamically. Compared with the previous SVM ensemble method and the traditional face identification method, the proposed SVM classification tree gets a very promising improvement in stability performance as the group size increases larger.
This communication presents a non-supervised restoration method based on a constrained Wiener filter. We implement our filter in the spatial domain and perform the filtering in 3-D. Our central contribution lies in th...
详细信息
This communication presents a non-supervised restoration method based on a constrained Wiener filter. We implement our filter in the spatial domain and perform the filtering in 3-D. Our central contribution lies in the derivation of a cross validation based algorithm which estimates the noise variance from the observed image. Exploitation of the partitioned matrix inversion lemma leads to a reasonable computation time. Results indicate that the method is able to determine the noise variance with an accuracy sufficient to produce acceptable results in the restoration at low signal-to-noise ratios. However at higher signal-to-noise ratios (above 15 dB) some undersmoothing is observed.
A recently-developed object segmentation algorithm in the work of Kim et al. (2006) achieved real-time generation of object masks with reasonable accuracy. However, this algorithm wastes a memory space and requires un...
详细信息
A recently-developed object segmentation algorithm in the work of Kim et al. (2006) achieved real-time generation of object masks with reasonable accuracy. However, this algorithm wastes a memory space and requires unnecessary computations because it uses an entire image frame to extract a video object. This paper modifies the segmentation algorithm according to Kim et al. (2006) to enable block-level processing of video object segmentation. An image frame is partitioned into 8times8 blocks, each of which is processed independently with the segmentation information inherited from its left and upper neighboring blocks as well as the previous frame. Simulation results show that the block-level processing algorithm saves 36% of the memory size and speeds up the computation by 361%. As a result, the proposed algorithm can process QCIF-size video at over 20 fps with ARM7TDMI running at 100 MHz.
In recent years we witnessed the explosion in the availability of data regarding human and customer behavior in the market. This data richness era has fostered the development of useful applications in understanding h...
详细信息
In recent years we witnessed the explosion in the availability of data regarding human and customer behavior in the market. This data richness era has fostered the development of useful applications in understanding how markets and the minds of the customers work. In this paper we focus on the analysis of complex networks based on customer behavior. Complex network analysis has provided a new and wide toolbox for the classic data mining task of clustering. With community discovery, i.e. the detection of functional modules in complex networks, we are now able to group together customers and products using a variety of different criteria. The aim of this paper is to explore this new analytic degree of freedom. We are interested in providing a case study uncovering the meaning of different community discovery algorithms on a network of products connected together because co-purchased by the same customers. We focus our interest in the different interpretation of a partition approach, where each product belongs to a single community, against an overlapping approach, where each product can belong to multiple communities. We found that the former is useful to improve the marketing classification of products, while the latter is able to create a collection of different customer profiles.
暂无评论