Owing to the flexible architectures of deep convolutional neural networks(CNNs)are successfully used for image denoising.However,they suffer from the following drawbacks:(i)deep network architecture is very difficult ...Owing to the flexible architectures of deep convolutional neural networks(CNNs)are successfully used for image denoising.However,they suffer from the following drawbacks:(i)deep network architecture is very difficult to train.(ii)Deeper networks face the challenge of performance saturation.In this study,the authors propose a novel method called enhanced convolutional neural denoising network(ECNDNet).Specifically,they use residual learning and batch normalisation techniques to address the problem of training difficulties and accelerate the convergence of the network.In addition,dilated convolutions are used in the proposed network to enlarge the context information and reduce the computational cost.Extensive experiments demonstrate that the ECNDNet outperforms the state-of-the-art methods for image denoising.展开更多
In recent years, instance-level-image retrieval has attracted massive attention. Several researchers proposed that the representations learned by convolutional neural network （CNN） can be used for image retrieval ta...In recent years, instance-level-image retrieval has attracted massive attention. Several researchers proposed that the representations learned by convolutional neural network （CNN） can be used for image retrieval task. In this study, the authors propose an effective feature encoder to extract robust information from CNN. It consists of two main steps： the embedding step and the aggregation step. Moreover, they apply the multi-task loss function to train their model in order to make the training process more effective. Finally, this study proposes a novel representation policy that encodes feature vectors extracted from different layers to capture both local patterns and semantic concepts from deep CNN. They call this ＇multi-level-image representation＇, which could further improve the performance. The proposed model is helpful to improve the retrieval performance. For the sake of comprehensively evaluating the performance of their approach, they conducted ablation experiments with various convolutional NN architectures. Furthermore, they apply their approach to a concrete challenge - Alibaba large-scale search challenge. The results show that their model is effective and competitive.展开更多
Foreground detection methods can be applied to efficiently distinguish foreground objects including moving or static objects from back- ground which is very important in the application of video analysis, especially v...Foreground detection methods can be applied to efficiently distinguish foreground objects including moving or static objects from back- ground which is very important in the application of video analysis, especially video surveillance. An excellent background model can obtain a good foreground detection results. A lot of background modeling methods had been proposed, but few comprehensive evaluations of them are available. These methods suffer from various challenges such as illumination changes and dynamic background. This paper first analyzed advantages and disadvantages of various background modeling methods in video analysis applications and then compared their performance in terms of quality and the computational cost. The Change detection.Net （CDnet2014） dataset and another video dataset with different envi- ronmental conditions （indoor, outdoor, snow） were used to test each method. The experimental results sufficiently demonstrated the strengths and drawbacks of traditional and recently proposed state-of-the-art background modeling methods. This work is helpful for both researchers and engineering practitioners. Codes of background modeling methods evaluated in this paper are available atwww.yongxu.org/lunwen.html.展开更多
Foley-Sammon linear discriminant analysis (FSLDA) and uncorrelated linear discriminant analysis (ULDA) are two well-known kinds of linear discriminant analysis. Both ULDA and FSLDA search the kth discriminant vector i...Foley-Sammon linear discriminant analysis (FSLDA) and uncorrelated linear discriminant analysis (ULDA) are two well-known kinds of linear discriminant analysis. Both ULDA and FSLDA search the kth discriminant vector in an n-k+1 dimensional subspace, while they are subject to their respective constraints. Evidenced by strict demonstration, it is clear that in essence ULDA vectors are the covariance-orthogonal vectors of the corresponding eigen-equation. So, the algorithms for the covariance-orthogonal vectors are equivalent to the original algorithm of ULDA, which is time-consuming. Also, it is first revealed that the Fisher criterion value of each FSLDA vector must be not less than that of the corresponding ULDA vector by theory analysis. For a discriminant vector, the larger its Fisher criterion value is, the more powerful in discriminability it is. So, for FSLDA vectors, corresponding to larger Fisher criterion values is an advantage. On the other hand, in general any two feature components extracted by FSLDA vectors are statistically correlated with each other, which may make the discriminant vectors set at a disadvantageous position. In contrast to FSLDA vectors, any two feature components extracted by ULDA vectors are statistically uncorrelated with each other. Two experiments on CENPARMI handwritten numeral database and ORL database are performed. The experimental results are consistent with the theory analysis on Fisher criterion values of ULDA vectors and FSLDA vectors. The experiments also show that the equivalent algorithm of ULDA, presented in this paper, is much more efficient than the original algorithm of ULDA, as the theory analysis expects. Moreover, it appears that if there is high statistical correlation between feature components extracted by FSLDA vectors, FSLDA will not perform well, in spite of larger Fisher criterion value owned by every FSLDA vector. However, when the average correlation coefficient of feature components extracted by FSLDA vectors is at a low level, the performance of FSLDA are comparable with ULDA.展开更多
Blind source extraction （BSE） is particularly at- tractive to solve blind signal mixture problems where only a few source signals are desired. Many existing BSE methods do not take into account the existence of nois...Blind source extraction （BSE） is particularly at- tractive to solve blind signal mixture problems where only a few source signals are desired. Many existing BSE methods do not take into account the existence of noise and can only work well in noise-free environments. In practice, the desired signal is often contaminated by additional noise. Therefore, we try to tackle the problem of noisy component extraction. The reference signal carries enough prior information to dis- tinguish the desired signal from signal mixtures. According to the useful properties of Gaussian moments, we incorporate the reference signal into a negentropy objective function so as to guide the extraction process and develop an improved BSE method. Extensive computer simulations demonstrate its validity in the process of revealing the underlying desired signal.展开更多
文摘Owing to the flexible architectures of deep convolutional neural networks(CNNs)are successfully used for image denoising.However,they suffer from the following drawbacks:(i)deep network architecture is very difficult to train.(ii)Deeper networks face the challenge of performance saturation.In this study,the authors propose a novel method called enhanced convolutional neural denoising network(ECNDNet).Specifically,they use residual learning and batch normalisation techniques to address the problem of training difficulties and accelerate the convergence of the network.In addition,dilated convolutions are used in the proposed network to enlarge the context information and reduce the computational cost.Extensive experiments demonstrate that the ECNDNet outperforms the state-of-the-art methods for image denoising.
文摘In recent years, instance-level-image retrieval has attracted massive attention. Several researchers proposed that the representations learned by convolutional neural network （CNN） can be used for image retrieval task. In this study, the authors propose an effective feature encoder to extract robust information from CNN. It consists of two main steps： the embedding step and the aggregation step. Moreover, they apply the multi-task loss function to train their model in order to make the training process more effective. Finally, this study proposes a novel representation policy that encodes feature vectors extracted from different layers to capture both local patterns and semantic concepts from deep CNN. They call this ＇multi-level-image representation＇, which could further improve the performance. The proposed model is helpful to improve the retrieval performance. For the sake of comprehensively evaluating the performance of their approach, they conducted ablation experiments with various convolutional NN architectures. Furthermore, they apply their approach to a concrete challenge - Alibaba large-scale search challenge. The results show that their model is effective and competitive.
文摘Foreground detection methods can be applied to efficiently distinguish foreground objects including moving or static objects from back- ground which is very important in the application of video analysis, especially video surveillance. An excellent background model can obtain a good foreground detection results. A lot of background modeling methods had been proposed, but few comprehensive evaluations of them are available. These methods suffer from various challenges such as illumination changes and dynamic background. This paper first analyzed advantages and disadvantages of various background modeling methods in video analysis applications and then compared their performance in terms of quality and the computational cost. The Change detection.Net （CDnet2014） dataset and another video dataset with different envi- ronmental conditions （indoor, outdoor, snow） were used to test each method. The experimental results sufficiently demonstrated the strengths and drawbacks of traditional and recently proposed state-of-the-art background modeling methods. This work is helpful for both researchers and engineering practitioners. Codes of background modeling methods evaluated in this paper are available atwww.yongxu.org/lunwen.html.
基金The National Natural Science Foundation of China (Grant No.60472060 ,60473039 and 60472061)
文摘Foley-Sammon linear discriminant analysis (FSLDA) and uncorrelated linear discriminant analysis (ULDA) are two well-known kinds of linear discriminant analysis. Both ULDA and FSLDA search the kth discriminant vector in an n-k+1 dimensional subspace, while they are subject to their respective constraints. Evidenced by strict demonstration, it is clear that in essence ULDA vectors are the covariance-orthogonal vectors of the corresponding eigen-equation. So, the algorithms for the covariance-orthogonal vectors are equivalent to the original algorithm of ULDA, which is time-consuming. Also, it is first revealed that the Fisher criterion value of each FSLDA vector must be not less than that of the corresponding ULDA vector by theory analysis. For a discriminant vector, the larger its Fisher criterion value is, the more powerful in discriminability it is. So, for FSLDA vectors, corresponding to larger Fisher criterion values is an advantage. On the other hand, in general any two feature components extracted by FSLDA vectors are statistically correlated with each other, which may make the discriminant vectors set at a disadvantageous position. In contrast to FSLDA vectors, any two feature components extracted by ULDA vectors are statistically uncorrelated with each other. Two experiments on CENPARMI handwritten numeral database and ORL database are performed. The experimental results are consistent with the theory analysis on Fisher criterion values of ULDA vectors and FSLDA vectors. The experiments also show that the equivalent algorithm of ULDA, presented in this paper, is much more efficient than the original algorithm of ULDA, as the theory analysis expects. Moreover, it appears that if there is high statistical correlation between feature components extracted by FSLDA vectors, FSLDA will not perform well, in spite of larger Fisher criterion value owned by every FSLDA vector. However, when the average correlation coefficient of feature components extracted by FSLDA vectors is at a low level, the performance of FSLDA are comparable with ULDA.
文摘Blind source extraction （BSE） is particularly at- tractive to solve blind signal mixture problems where only a few source signals are desired. Many existing BSE methods do not take into account the existence of noise and can only work well in noise-free environments. In practice, the desired signal is often contaminated by additional noise. Therefore, we try to tackle the problem of noisy component extraction. The reference signal carries enough prior information to dis- tinguish the desired signal from signal mixtures. According to the useful properties of Gaussian moments, we incorporate the reference signal into a negentropy objective function so as to guide the extraction process and develop an improved BSE method. Extensive computer simulations demonstrate its validity in the process of revealing the underlying desired signal.
