Bayesian Matting has four limitations.Firstly,Bayesian matting makes strong assumption that the texture distribution of nature image satisfies Gaussian distribution with fixed variance.This assumption will fail for co...Bayesian Matting has four limitations.Firstly,Bayesian matting makes strong assumption that the texture distribution of nature image satisfies Gaussian distribution with fixed variance.This assumption will fail for complex texture distribution.In order to extract the nature images with complex texture distribution,we design an information entropy approach to estimate the scalable variance.Secondly,when the opacity is near the boundary of the value range,Bayesian matting method may be failure because of the error computation of opacity.Therefore,a rectification approach is proposed to adjust the computation model and keep the opacity within the valid value range.Thirdly,Bayesian matting is a local sample method which may miss some valid samples of matting.We propose a selection function to integrate valid global sample matting result into above matting framework as a supplement to the local sample matting result.The proposed function is compose of three criteria,that is,the similarity of results,the overlapping degree of samples,and the similarity of neighborhood.Fourthly,in order to obtain a smooth and vivid matte,the result is further refined by considering correlation between neighbouring pixels.Finally,We use online benchmark for image matting to evaluate the proposed method with both qualitative observation and quantitative analysis.The experiments show that our method achieves a competitive advantages over other methods.展开更多
In recent years,there are numerous works been proposed to leverage the techniques of deep learning to improve social-aware recommendation performance.In most cases,it requires a larger number of data to train a robust...In recent years,there are numerous works been proposed to leverage the techniques of deep learning to improve social-aware recommendation performance.In most cases,it requires a larger number of data to train a robust deep learning model,which contains a lot of parameters to fit training data.However,both data of user ratings and social networks are facing critical sparse problem,which makes it not easy to train a robust deep neural network model.Towards this problem,we propose a novel correlative denoising autoencoder(CoDAE)method by taking correlations between users with multiple roles into account to learn robust representations from sparse inputs of ratings and social networks for recommendation.We develop the CoDAE model by utilizing three separated autoencoders to learn user features with roles of rater,truster and trustee,respectively.Especially,on account of that each input unit of user vectors with roles of truster and trustee is corresponding to a particular user,we propose to utilize shared parameters to learn common information of the units that corresponding to same users.Moreover,we propose a related regularization term to learn correlations between user features that learnt by the three subnetworks of CoDAE model.We further conduct a series of experiments to evaluate the proposed method on two public datasets for Top-N recommendation task.The experimental results demonstrate that the proposed model outperforms state-of-the-art algorithms on rank-sensitive metrics of MAP and NDCG.展开更多
文摘Bayesian Matting has four limitations.Firstly,Bayesian matting makes strong assumption that the texture distribution of nature image satisfies Gaussian distribution with fixed variance.This assumption will fail for complex texture distribution.In order to extract the nature images with complex texture distribution,we design an information entropy approach to estimate the scalable variance.Secondly,when the opacity is near the boundary of the value range,Bayesian matting method may be failure because of the error computation of opacity.Therefore,a rectification approach is proposed to adjust the computation model and keep the opacity within the valid value range.Thirdly,Bayesian matting is a local sample method which may miss some valid samples of matting.We propose a selection function to integrate valid global sample matting result into above matting framework as a supplement to the local sample matting result.The proposed function is compose of three criteria,that is,the similarity of results,the overlapping degree of samples,and the similarity of neighborhood.Fourthly,in order to obtain a smooth and vivid matte,the result is further refined by considering correlation between neighbouring pixels.Finally,We use online benchmark for image matting to evaluate the proposed method with both qualitative observation and quantitative analysis.The experiments show that our method achieves a competitive advantages over other methods.
基金supported by the National Natural Science Foundation of China(Grant No.61472289)the National Key Research and Development Project(2016YFC0106305).
文摘In recent years,there are numerous works been proposed to leverage the techniques of deep learning to improve social-aware recommendation performance.In most cases,it requires a larger number of data to train a robust deep learning model,which contains a lot of parameters to fit training data.However,both data of user ratings and social networks are facing critical sparse problem,which makes it not easy to train a robust deep neural network model.Towards this problem,we propose a novel correlative denoising autoencoder(CoDAE)method by taking correlations between users with multiple roles into account to learn robust representations from sparse inputs of ratings and social networks for recommendation.We develop the CoDAE model by utilizing three separated autoencoders to learn user features with roles of rater,truster and trustee,respectively.Especially,on account of that each input unit of user vectors with roles of truster and trustee is corresponding to a particular user,we propose to utilize shared parameters to learn common information of the units that corresponding to same users.Moreover,we propose a related regularization term to learn correlations between user features that learnt by the three subnetworks of CoDAE model.We further conduct a series of experiments to evaluate the proposed method on two public datasets for Top-N recommendation task.The experimental results demonstrate that the proposed model outperforms state-of-the-art algorithms on rank-sensitive metrics of MAP and NDCG.
