摘要 :
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this pap...
展开
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this paper. Firstly, the registration parameters based on affine model is initially got by FFT-based method. Secondly, we judge the result of the FFT-based registration by fitness value like through image correlation, if the fitness value is less then some threshold, we believe that the result is good enough and return the parameters of affine model; otherwise a GA optimization procedure is started with the initialization of FFT-based result. The GA optimization can converge much more fast because the initialization of FFT-based result is more close to the converge point than the random initialization. Experimental data show this method works well.
收起
摘要 :
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this pap...
展开
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this paper. Firstly, the registration parameters based on affine model is initially got by FFT-based method. Secondly, we judge the result of the FFT-based registration by fitness value like through image correlation, if the fitness value is less then some threshold, we believe that the result is good enough and return the parameters of affine model; otherwise a GA optimization procedure is started with the initialization of FFT-based result. The GA optimization can converge much more fast because the initialization of FFT-based result is more close to the converge point than the random initialization. Experimental data show this method works well.
收起
摘要 :
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this pap...
展开
Image registration is an important subject in computer vision. How to improve registration precision is a hard task. A feature-less method of registration based on GA Optimization with FFT Initialization is put forward in this paper. Firstly, the registration parameters based on affine model is initially got by FFT-based method. Secondly, we judge the result of the FFT-based registration by fitness value like through image correlation, if the fitness value is less then some threshold, we believe that the result is good enough and return the parameters of affine model; otherwise a GA optimization procedure is started with the initialization of FFT-based result. The GA optimization can converge much more fast because the initialization of FFT-based result is more close to the converge point than the random initialization. Experimental data show this method works well.
收起
摘要 :
Some finished and ongoing projects of digital music library are introduced in this paper, the structure and related technologies are discussed, and our approach of content-based music retrieval is also presented.
摘要 :
Some finished and ongoing projects of digital music library are introduced in this paper, the structure and related technologies are discussed, and our approach of content-based music retrieval is also presented.
摘要 :
This paper describes the design, implementation and usage of a secure scalable Byzantine fault tolerant MDS system in the Grid. The scalable Byzantine fault tolerant MDS system provides a hierarchy GIIS servers, a local GIIS domai...
展开
This paper describes the design, implementation and usage of a secure scalable Byzantine fault tolerant MDS system in the Grid. The scalable Byzantine fault tolerant MDS system provides a hierarchy GIIS servers, a local GIIS domain can require the resource it needs from remote GIIS domain. By using the state-machine replication approach and quorum system technique, the scalable Byzantine fault tolerant MDS system can tolerate not only benign faults, but also arbitrary (Byzantine) ones. An optimizing key management and caching mechanism is also explored in this paper.
收起
摘要 :
We present in this paper a novel system for query by humming, our method differs from other ones in the followings: Firstly, we use recurrent neural network as the index of music database. Secondly, we present correlation degree t...
展开
We present in this paper a novel system for query by humming, our method differs from other ones in the followings: Firstly, we use recurrent neural network as the index of music database. Secondly, we present correlation degree to evaluate the music matching precision. We now hold a database of 201 pieces of music with various genres. The result of our experiment reports that the successful rate is 63% with top one matching and 87% with top three matching. Future work will be on melody extraction technique from popular formats of music and on-line music retrieval.
收起
摘要 :
Practice is the key step in training softwareoperation skills. Unfortunately, teachers' Grading of operational assignments is a laborious and tedious task, a problem also accompanied by numerous urge for the real time instruction ...
展开
Practice is the key step in training softwareoperation skills. Unfortunately, teachers' Grading of operational assignments is a laborious and tedious task, a problem also accompanied by numerous urge for the real time instruction from the students. Our solution to this problem is to automate the processes of assignment grading and knowledge-level evaluating such that students can receive tailored feedback instantly. This paper presents the architecture and mechanism of the adaptive training system, and generalizes a universal approach for grading operational assignments. In addition, for diagnosing uncertain knowledge level of students, Bayesian Networks and Computer Adaptive Tests (CAT) technologies were adopted. This system has been successfully employed in hundreds of universities and schools. The results obtained from the educational activities indicate its excellent performance in both efficiency and accuracy.
收起
摘要 :
Practice is the key step in training softwareoperation skills. Unfortunately, teachers' Grading of operational assignments is a laborious and tedious task, a problem also accompanied by numerous urge for the real time instruction ...
展开
Practice is the key step in training softwareoperation skills. Unfortunately, teachers' Grading of operational assignments is a laborious and tedious task, a problem also accompanied by numerous urge for the real time instruction from the students. Our solution to this problem is to automate the processes of assignment grading and knowledge-level evaluating such that students can receive tailored feedback instantly. This paper presents the architecture and mechanism of the adaptive training system, and generalizes a universal approach for grading operational assignments. In addition, for diagnosing uncertain knowledge level of students, Bayesian Networks and Computer Adaptive Tests (CAT) technologies were adopted. This system has been successfully employed in hundreds of universities and schools. The results obtained from the educational activities indicate its excellent performance in both efficiency and accuracy.
收起
摘要 :
In video based human animation with two cameras, the result of feature tracking affects the vividness of reconstructed 3D human motion. So a good tracking algorithm is able to improve the visual effect of animation greatly. This p...
展开
In video based human animation with two cameras, the result of feature tracking affects the vividness of reconstructed 3D human motion. So a good tracking algorithm is able to improve the visual effect of animation greatly. This paper presents a new tracking algorithm, which uses Kalman filter to predict the image coordinates of joints, then uses subblock-based model to match feature point in the fore-and-aft frames, and uses epipolar line equation to guide tracking and the feature point with big error is ruled out. In the world coordinate system, estimating the feature points by Kalman filter algorithm can solve the occlusion problem. The presented tracking algorithm has been applied to VBHA system implemented by our lab. Experimental results prove that the speed and efficiency of this algorithm can meet the requirement of video-based animation.
收起