Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553519
Amelia Villegas-Morcillo, J. A. Morales-Cordovilla, A. Gómez, V. Sánchez
A protein contact map is a simplified matrix representation of the protein structure, where the spatial proximity of two amino acid residues is reflected. Although the accurate prediction of protein inter-residue contacts from the amino acid sequence is an open problem, considerable progress has been made in recent years. This progress has been driven by the development of contact predictors that identify the coevolutionary events occurring in a protein multiple sequence alignment (MSA). However, it has been shown that these methods introduce Gaussian noise in the estimated contact map, making its reduction necessary. In this paper, we propose the use of two different Gaussian denoising approximations in order to enhance the protein contact estimation. These approaches are based on (i) sparse representations over learned dictionaries, and (ii) deep residual convolutional neural networks. The results highlight that the residual learning strategy allows a better reconstruction of the contact map, thus improving contact predictions.
{"title":"Improved Protein Residue-Residue Contact Prediction Using Image Denoising Methods","authors":"Amelia Villegas-Morcillo, J. A. Morales-Cordovilla, A. Gómez, V. Sánchez","doi":"10.23919/EUSIPCO.2018.8553519","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553519","url":null,"abstract":"A protein contact map is a simplified matrix representation of the protein structure, where the spatial proximity of two amino acid residues is reflected. Although the accurate prediction of protein inter-residue contacts from the amino acid sequence is an open problem, considerable progress has been made in recent years. This progress has been driven by the development of contact predictors that identify the coevolutionary events occurring in a protein multiple sequence alignment (MSA). However, it has been shown that these methods introduce Gaussian noise in the estimated contact map, making its reduction necessary. In this paper, we propose the use of two different Gaussian denoising approximations in order to enhance the protein contact estimation. These approaches are based on (i) sparse representations over learned dictionaries, and (ii) deep residual convolutional neural networks. The results highlight that the residual learning strategy allows a better reconstruction of the contact map, thus improving contact predictions.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121416838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553576
Thi-Thu-Hong Phan, É. Poisson, A. Bigand
Time series forecasting has an important role in many real applications in meteorology and environment to understand phenomena as climate change and to adapt monitoring strategy. This paper aims first to build a framework for forecasting meteorological univariate time series and then to carry out a performance comparison of different univariate models for forecasting task. Six algorithms are discussed: Single exponential smoothing (SES), Seasonal-naive (Snaive), Seasonal-ARIMA (SARIMA), Feed-Forward Neural Network (FFNN), Dynamic Time Warping-based Imputation (DTWBI), Bayesian Structural Time Series (BSTS). Four performance measures and various meteorological time series are used to determine a more customized method for forecasting. Through experiments results, FFNN method is well adapted to forecast meteorological univariate time series with seasonality and no trend in consideration of accuracy indices and DTWBI is more suitable as considering the shape and dynamics of forecast values.
{"title":"Comparative Study on Univariate Forecasting Methods for Meteorological Time Series","authors":"Thi-Thu-Hong Phan, É. Poisson, A. Bigand","doi":"10.23919/EUSIPCO.2018.8553576","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553576","url":null,"abstract":"Time series forecasting has an important role in many real applications in meteorology and environment to understand phenomena as climate change and to adapt monitoring strategy. This paper aims first to build a framework for forecasting meteorological univariate time series and then to carry out a performance comparison of different univariate models for forecasting task. Six algorithms are discussed: Single exponential smoothing (SES), Seasonal-naive (Snaive), Seasonal-ARIMA (SARIMA), Feed-Forward Neural Network (FFNN), Dynamic Time Warping-based Imputation (DTWBI), Bayesian Structural Time Series (BSTS). Four performance measures and various meteorological time series are used to determine a more customized method for forecasting. Through experiments results, FFNN method is well adapted to forecast meteorological univariate time series with seasonality and no trend in consideration of accuracy indices and DTWBI is more suitable as considering the shape and dynamics of forecast values.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"448 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116758868","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553131
P. Baggenstoss
Real-world acoustic events span a wide range of time and frequency resolutions, from short clicks to longer tonals. This is a challenge for the hidden Markov model (HMM), which uses a fixed segmentation and feature extraction, forcing a compromise between time and frequency resolution. The multiresolution HMM (MR-HMM) is an extension of the HMM that assumes not only an underlying (hidden) random state sequence, but also an underlying random segmentation, with segments spanning a wide range of sizes and processed using a variety of feature extraction methods. It is shown that the MR-HMM alone, as an acoustic event classifier, has performance comparable to state of the art discriminative classifiers on three open data sets. However, as a generative classifier, the MR-HMM models the underlying data generation process and can generate synthetic data, allowing weaknesses of individual class models to be discovered and corrected. To demonstrate this point, the MR-HMM is combined with auxiliary features that capture temporal information, resulting in significantly improved performance.
{"title":"Acoustic Event Classification Using Multi-Resolution HMM","authors":"P. Baggenstoss","doi":"10.23919/EUSIPCO.2018.8553131","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553131","url":null,"abstract":"Real-world acoustic events span a wide range of time and frequency resolutions, from short clicks to longer tonals. This is a challenge for the hidden Markov model (HMM), which uses a fixed segmentation and feature extraction, forcing a compromise between time and frequency resolution. The multiresolution HMM (MR-HMM) is an extension of the HMM that assumes not only an underlying (hidden) random state sequence, but also an underlying random segmentation, with segments spanning a wide range of sizes and processed using a variety of feature extraction methods. It is shown that the MR-HMM alone, as an acoustic event classifier, has performance comparable to state of the art discriminative classifiers on three open data sets. However, as a generative classifier, the MR-HMM models the underlying data generation process and can generate synthetic data, allowing weaknesses of individual class models to be discovered and corrected. To demonstrate this point, the MR-HMM is combined with auxiliary features that capture temporal information, resulting in significantly improved performance.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"107 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122431050","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553544
Coralie Martinez, G. Perrin, E. Ramasso, M. Rombaut
In many real-world applications, ranging from predictive maintenance to personalized medicine, early classification of time series data is of paramount importance for supporting decision makers. In this article, we address this challenging task with a novel approach based on reinforcement learning. We introduce an early classifier agent, an end-to-end reinforcement learning agent (deep Q-network, DQN) [1] able to perform early classification in an efficient way. We formulate the early classification problem in a reinforcement learning framework: we introduce a suitable set of states and actions but we also define a specific reward function which aims at finding a compromise between earliness and classification accuracy. While most of the existing solutions do not explicitly take time into account in the final decision, this solution allows the user to set this trade-off in a more flexible way. In particular, we show experimentally on datasets from the UCR time series archive [2] that this agent is able to continually adapt its behavior without human intervention and progressively learn to compromise between accurate and fast predictions.
{"title":"A Deep Reinforcement Learning Approach for Early Classification of Time Series","authors":"Coralie Martinez, G. Perrin, E. Ramasso, M. Rombaut","doi":"10.23919/EUSIPCO.2018.8553544","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553544","url":null,"abstract":"In many real-world applications, ranging from predictive maintenance to personalized medicine, early classification of time series data is of paramount importance for supporting decision makers. In this article, we address this challenging task with a novel approach based on reinforcement learning. We introduce an early classifier agent, an end-to-end reinforcement learning agent (deep Q-network, DQN) [1] able to perform early classification in an efficient way. We formulate the early classification problem in a reinforcement learning framework: we introduce a suitable set of states and actions but we also define a specific reward function which aims at finding a compromise between earliness and classification accuracy. While most of the existing solutions do not explicitly take time into account in the final decision, this solution allows the user to set this trade-off in a more flexible way. In particular, we show experimentally on datasets from the UCR time series archive [2] that this agent is able to continually adapt its behavior without human intervention and progressively learn to compromise between accurate and fast predictions.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122588802","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553496
Ioan-Catalin Dragoi, D. Coltuc
Pairwise reversible data hiding (RDH) restricts the embedding to 3 combinations of bits per pixel pair (“00”, “01”, “10”), by eliminating the embedding of “1” into both pixels. The gain in quality is significant and the loss in embedding bitrate is compensated by embedding into previously shifted pairs. This restriction requires a special coding procedure to format the encrypted hidden data. This paper proposes a new set of embedding equations for pairwise RDH. The proposed approach inserts either one or two data bits into each pair based on its type, bypassing the need for special coding. The proposed equations can be easily integrated in most pairwise reversible data hiding frameworks. They also provide more room for data embedding than their classic counterparts at the low embedding distortion required for high-fidelity RDH.
{"title":"Improved Pairwise Embedding for High-Fidelity Reversible Data Hiding","authors":"Ioan-Catalin Dragoi, D. Coltuc","doi":"10.23919/EUSIPCO.2018.8553496","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553496","url":null,"abstract":"Pairwise reversible data hiding (RDH) restricts the embedding to 3 combinations of bits per pixel pair (“00”, “01”, “10”), by eliminating the embedding of “1” into both pixels. The gain in quality is significant and the loss in embedding bitrate is compensated by embedding into previously shifted pairs. This restriction requires a special coding procedure to format the encrypted hidden data. This paper proposes a new set of embedding equations for pairwise RDH. The proposed approach inserts either one or two data bits into each pair based on its type, bypassing the need for special coding. The proposed equations can be easily integrated in most pairwise reversible data hiding frameworks. They also provide more room for data embedding than their classic counterparts at the low embedding distortion required for high-fidelity RDH.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122674515","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553507
V. Bruni, L. D. Cioppa, D. Vitulano
In this paper an automatic method for the selection of those Fourier descriptors which better correlate a 2D shape contour is presented. To this aim, shape description has been modeled as a non linear approximation problem and a strict relationship between transform entropy and the sorted version of the transformed analysed boundary is derived. As a result, Fourier descriptors are selected in a hierarchical way and the minimum number of coefficients able to give a nearly optimal shape boundary representation is automatically derived. The latter maximizes an entropic interpretation of a complexity-based similarity measure, i.e. the normalized information distance. Preliminary experimental results show that the proposed method is able to provide a compact and computationally effective description of shape boundary which guarantees a nearly optimal matching with the original one.
{"title":"An entropy-based approach for shape description","authors":"V. Bruni, L. D. Cioppa, D. Vitulano","doi":"10.23919/EUSIPCO.2018.8553507","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553507","url":null,"abstract":"In this paper an automatic method for the selection of those Fourier descriptors which better correlate a 2D shape contour is presented. To this aim, shape description has been modeled as a non linear approximation problem and a strict relationship between transform entropy and the sorted version of the transformed analysed boundary is derived. As a result, Fourier descriptors are selected in a hierarchical way and the minimum number of coefficients able to give a nearly optimal shape boundary representation is automatically derived. The latter maximizes an entropic interpretation of a complexity-based similarity measure, i.e. the normalized information distance. Preliminary experimental results show that the proposed method is able to provide a compact and computationally effective description of shape boundary which guarantees a nearly optimal matching with the original one.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131564243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553084
Rodrigo Cabral Farias, J. H. D. M. Goulart, P. Comon
In this paper we present a modification of alternating least squares (ALS) for tensor canonical polyadic approximation that takes into account mutual coherence constraints. The proposed algorithm can be used to ensure well-posedness of the tensor approximation problem during ALS iterates and so is an alternative to existing approaches. We conduct tests with the proposed approach by using it as initialization of unconstrained alternating least squares in difficult cases, when the underlying tensor model factors have nearly collinear columns and the unconstrained approach is prone to a degenerate behavior, failing to converge or converging slowly to an acceptable solution. The results of the tested cases indicate that by using such an initialization the unconstrained approach seems to avoid such a behavior.
{"title":"Coherence Constrained Alternating Least Squares","authors":"Rodrigo Cabral Farias, J. H. D. M. Goulart, P. Comon","doi":"10.23919/EUSIPCO.2018.8553084","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553084","url":null,"abstract":"In this paper we present a modification of alternating least squares (ALS) for tensor canonical polyadic approximation that takes into account mutual coherence constraints. The proposed algorithm can be used to ensure well-posedness of the tensor approximation problem during ALS iterates and so is an alternative to existing approaches. We conduct tests with the proposed approach by using it as initialization of unconstrained alternating least squares in difficult cases, when the underlying tensor model factors have nearly collinear columns and the unconstrained approach is prone to a degenerate behavior, failing to converge or converging slowly to an acceptable solution. The results of the tested cases indicate that by using such an initialization the unconstrained approach seems to avoid such a behavior.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132357520","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553138
Sergio Moreschini, G. Scrofani, R. Bregović, G. Saavedra, A. Gotchev
Integral or light field imaging is an attractive approach in microscopy, as it allows to capture 3D samples in just one shot and explore them later through changing the focus on particular depth planes of interest. However, it requires a compromise between spatial and angular resolution on the 2D sensor recording the microscopic images. A particular setting called Fourier Integral Microscope (FIMic) allows maximizing the spatial resolution for the cost of reducing the angular one. In this work, we propose a technique, which aims at reconstructing the continuous light field from sparse FIMic measurements, thus providing the functionality of continuous refocus on any arbitrary depth plane. Our main tool is the densely-sampled light field reconstruction in shearlet domain specifically tailored for the case of FIMic. The experiments demonstrate that the implemented technique yields better results compared to refocusing sparsely-sampled data.
{"title":"Continuous Refocusing for Integral Microscopy with Fourier Plane Recording","authors":"Sergio Moreschini, G. Scrofani, R. Bregović, G. Saavedra, A. Gotchev","doi":"10.23919/EUSIPCO.2018.8553138","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553138","url":null,"abstract":"Integral or light field imaging is an attractive approach in microscopy, as it allows to capture 3D samples in just one shot and explore them later through changing the focus on particular depth planes of interest. However, it requires a compromise between spatial and angular resolution on the 2D sensor recording the microscopic images. A particular setting called Fourier Integral Microscope (FIMic) allows maximizing the spatial resolution for the cost of reducing the angular one. In this work, we propose a technique, which aims at reconstructing the continuous light field from sparse FIMic measurements, thus providing the functionality of continuous refocus on any arbitrary depth plane. Our main tool is the densely-sampled light field reconstruction in shearlet domain specifically tailored for the case of FIMic. The experiments demonstrate that the implemented technique yields better results compared to refocusing sparsely-sampled data.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"99 1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128007149","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553183
Najmeddine Majed, S. Ragot, Lætitia Gros, X. Lagrange, Alberto Blanc
In this paper, we study the performance of the 3GPP EVS codec when this codec is used in conjunction with 100% application-layer redundancy. The objective of this work is to investigate potential performance gains for Voice over LTE (VoLTE) in bad coverage scenarios. Voice quality for the EVS codec operated in the 9.6-24.4 kbit/s bit range in super-wideband (SWB) is evaluated at different packet loss rates (PLR), using objective and subjective methods (iTu - T P.863 and P.800 ACR). Results show that EVS at 9.6 kbit/s with 100% application-layer redundancy has significantly higher packet loss resilience in degraded channel conditions (≥ 3 % PLR), for an overall bit rate (around 2×9.6 kbit/s) compatible with VoLTE (assuming a VoLTE bearer configured to a maximum rate of 24.4 kbit/s). We also discuss the relative merit of the partial redundancy mode in the EVS codec at 13.2 kbit/s, known as the channel-aware mode (CAM), and possible RTP/RTCP signaling methods to trigger the use of application-layer redundancy.
{"title":"Application-Layer Redundancy for the EVS Codec","authors":"Najmeddine Majed, S. Ragot, Lætitia Gros, X. Lagrange, Alberto Blanc","doi":"10.23919/EUSIPCO.2018.8553183","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553183","url":null,"abstract":"In this paper, we study the performance of the 3GPP EVS codec when this codec is used in conjunction with 100% application-layer redundancy. The objective of this work is to investigate potential performance gains for Voice over LTE (VoLTE) in bad coverage scenarios. Voice quality for the EVS codec operated in the 9.6-24.4 kbit/s bit range in super-wideband (SWB) is evaluated at different packet loss rates (PLR), using objective and subjective methods (iTu - T P.863 and P.800 ACR). Results show that EVS at 9.6 kbit/s with 100% application-layer redundancy has significantly higher packet loss resilience in degraded channel conditions (≥ 3 % PLR), for an overall bit rate (around 2×9.6 kbit/s) compatible with VoLTE (assuming a VoLTE bearer configured to a maximum rate of 24.4 kbit/s). We also discuss the relative merit of the partial redundancy mode in the EVS codec at 13.2 kbit/s, known as the channel-aware mode (CAM), and possible RTP/RTCP signaling methods to trigger the use of application-layer redundancy.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132721927","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-09-01DOI: 10.23919/EUSIPCO.2018.8553082
Zhi Jin, Naili Luo, Lei Luo, Wenbin Zou, Xia Li, E. Steinbach
Stereo images provide users with a vivid 3D watching experience. Supported by per-view depth maps, 3D stereo images can be used to generate any required intermediate view between the given left and right stereo views. However, 3D stereo images lead to higher transmission and storage cost compared to single view images. Based on the binocular suppression theory, mixed-quality stereo images can alleviate this problem by employing different compression ratios on the two views. This causes noticeable visual artifacts when a high compression ratio is adopted and limits free-viewpoint applications. Hence, the low quality image at the receiver side needs to be enhanced to match the high quality one. To address this problem, in this paper we propose an end-to-end fully Convolutional Neural Network (CNN) for enhancing the low quality images in quality asymmetric stereo images by exploiting inter-view correlation. The proposed network achieves an image quality boost of up to 4.6dB and 3.88dB PSNR gain over ordinary JPEG for QF10 and 20, respectively, and an improvement of up to 2.37dB and 2.05dB over the state-of-the-art CNN-based results for QF10 and 20, respectively.
{"title":"Information Fusion based Quality Enhancement for 3D Stereo Images Using CNN","authors":"Zhi Jin, Naili Luo, Lei Luo, Wenbin Zou, Xia Li, E. Steinbach","doi":"10.23919/EUSIPCO.2018.8553082","DOIUrl":"https://doi.org/10.23919/EUSIPCO.2018.8553082","url":null,"abstract":"Stereo images provide users with a vivid 3D watching experience. Supported by per-view depth maps, 3D stereo images can be used to generate any required intermediate view between the given left and right stereo views. However, 3D stereo images lead to higher transmission and storage cost compared to single view images. Based on the binocular suppression theory, mixed-quality stereo images can alleviate this problem by employing different compression ratios on the two views. This causes noticeable visual artifacts when a high compression ratio is adopted and limits free-viewpoint applications. Hence, the low quality image at the receiver side needs to be enhanced to match the high quality one. To address this problem, in this paper we propose an end-to-end fully Convolutional Neural Network (CNN) for enhancing the low quality images in quality asymmetric stereo images by exploiting inter-view correlation. The proposed network achieves an image quality boost of up to 4.6dB and 3.88dB PSNR gain over ordinary JPEG for QF10 and 20, respectively, and an improvement of up to 2.37dB and 2.05dB over the state-of-the-art CNN-based results for QF10 and 20, respectively.","PeriodicalId":303069,"journal":{"name":"2018 26th European Signal Processing Conference (EUSIPCO)","volume":"9 2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134352223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：