Advances in multimedia information processing PCM 2017 part i

LNCS 10735 Bing Zeng · Qingming Huang Abdulmotaleb El Saddik · Hongliang Li Shuqiang Jiang · Xiaopeng Fan (Eds.) Advances in Multimedia Information Processing – PCM 2017 18th Pacific-Rim Conference on Multimedia Harbin, China, September 28–29, 2017 Revised Selected Papers, Part I 123 Lecture Notes in Computer Science Commenced Publication in 1973 Founding and Former Series Editors: Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen Editorial Board David Hutchison Lancaster University, Lancaster, UK Takeo Kanade Carnegie Mellon University, Pittsburgh, PA, USA Josef Kittler University of Surrey, Guildford, UK Jon M Kleinberg Cornell University, Ithaca, NY, USA Friedemann Mattern ETH Zurich, Zurich, Switzerland John C Mitchell Stanford University, Stanford, CA, USA Moni Naor Weizmann Institute of Science, Rehovot, Israel C Pandu Rangan Indian Institute of Technology Madras, Chennai, India Bernhard Steffen TU Dortmund University, Dortmund, Germany Demetri Terzopoulos University of California, Los Angeles, CA, USA Doug Tygar University of California, Berkeley, CA, USA Gerhard Weikum Max Planck Institute for Informatics, Saarbrücken, Germany 10735 More information about this series at http://www.springer.com/series/7409 Bing Zeng Qingming Huang Abdulmotaleb El Saddik Hongliang Li Shuqiang Jiang Xiaopeng Fan (Eds.) • • • Advances in Multimedia Information Processing – PCM 2017 18th Pacific-Rim Conference on Multimedia Harbin, China, September 28–29, 2017 Revised Selected Papers, Part I 123 Editors Bing Zeng University of Electronic Science and Technology of China Chengdu China Hongliang Li University of Electronic Science and Technology of China Chengdu China Qingming Huang University of Chinese Academy of Sciences Beijing China Shuqiang Jiang Chinese Academy of Sciences Beijing China Abdulmotaleb El Saddik University of Ottawa Ottawa, ON Canada Xiaopeng Fan Harbin Institute of Technology Harbin China ISSN 0302-9743 ISSN 1611-3349 (electronic) Lecture Notes in Computer Science ISBN 978-3-319-77379-7 ISBN 978-3-319-77380-3 (eBook) https://doi.org/10.1007/978-3-319-77380-3 Library of Congress Control Number: 2018935899 LNCS Sublibrary: SL3 – Information Systems and Applications, incl Internet/Web, and HCI © Springer International Publishing AG, part of Springer Nature 2018 This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations Printed on acid-free paper This Springer imprint is published by the registered company Springer International Publishing AG part of Springer Nature The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland Preface On behalf of the Organizing Committee, it is our great pleasure to welcome you to the proceedings of the 2017 Pacific-Rim Conference on Multimedia (PCM 2017) PCM serves as an international forum to bring together researchers and practitioners from academia and industry to discuss research on state-of-the-art Internet multimedia processing, multimedia service, analysis, and applications PCM 2017 was the 18th in the series that has been held annually since 2000 In 2017, PCM was held in Harbin, China Consistent with previous editions of PCM, we prepared a very attractive technical program with two keynote talks, one best paper candidate session, nine oral presentation sessions, two poster sessions, and six oral special sessions Moreover, thanks to the co-organization with IEEE CAS Beijing chapter, this year’s program featured a panel session titled “Advanced Multimedia Technology.” Social and intellectual interactions were enjoyed among students, young researchers, and leading scholars We received 264 submissions for regular papers this year These submissions cover the areas of multimedia content analysis, multimedia signal processing and systems, multimedia applications and services, etc We thank our 104 Technical Program Committee members for their efforts in reviewing papers and providing valuable feedback to the authors From the total of 264 submissions and based on at least two reviews per submission, the Program Chairs decided to accept 48 oral papers (18.2%) and 96 poster papers, i.e, the overall acceptance ratio for regular paper is 54.9% Among the 48 oral papers, two papers received the Best Paper and the Best Student Paper award Moreover, we accepted six special sessions with 35 papers The technical program is an important aspect but only delivers its full impact if surrounded by challenging keynotes We are extremely pleased and grateful to have two exceptional keynote speakers, Wenwu Zhu and Josep Lladós, accept our invitation and present interesting ideas and insights at PCM 2017 We would also like to express our sincere gratitude to all the other Organizing Committee members, the general chairs, Bing Zeng, Qingming Huang, and Abdulmotaleb El Saddik, the program chair, Hongliang Li, Shuqiang Jiang, and Xiaopeng Fan, the panel chairs, Zhu Li and Debin Zhao, the organizing chairs, Shaohui Liu, Liang Li, and Yan Chen, the publication chairs, Shuhui Wang and Wen-Huang Cheng, the sponsorship chairs, Wangmeng Zuo, Luhong Liang, and Ke Lv, the registration and finance chairs, Guorong Li and Weiqing Min, among others Their outstanding effort contributed to this extremely rich and complex main program that characterizes PCM 2017 Last but not the least, we thank VI Preface all the authors, session chairs, student volunteers, and supporters Their contributions are much appreciated We sincerely hope that you will enjoy reading the proceedings of PCM 2017 September 2017 Bing Zeng Qingming Huang Abdulmotaleb El Saddik Hongliang Li Shuqiang Jiang Xiaopeng Fan Organization Organizing Committee General Chairs Bing Zeng Qingming Huang Abdulmotaleb El Saddik University of Electronic Science and Technology of China University of Chinese Academy of Sciences, China University of Ottawa, Canada Program Chairs Hongliang Li Shuqiang Jiang Xiaopeng Fan University of Electronic Science and Technology of China ICT, Chinese Academy of Sciences, China Harbin Institute of Technology, China Organizing Chairs Shaohui Liu Liang Li Yan Chen Harbin Institute of Technology, China University of Chinese Academy Sciences, China University of Electronic Science and Technology of China Panel Chairs Zhu Li Debin Zhao University of Missouri-Kansas City, USA Harbin Institute of Technology, China Technical Committee Publication Chairs Shuhui Wang Wen-Huang Cheng Tongwei Ren Lu Fang ICT, Chinese Academy of Sciences, China Taiwan Academia Sinica, Taiwan Nanjing University, China Hong Kong University of Science and Technology, SAR China Special Session Chairs Yan Liu Yu-Gang Jiang Wen Ji Jinqiao Shi Feng Jiang The Hong Kong Polytechnic University, SAR China Fudan University, China ICT, Chinese Academy of Sciences, China Chinese Academy Sciences, China Harbin Institute of Technology, China VIII Organization Tutorial Chairs Zheng-jun Zha Siwei Ma Chong-Wah Ngo Ruiqin Xiong Hefei Institute of Intelligent Machines, Chinese Academy of Sciences, China Peking University, China City University of Hong Kong, SAR China Peking University, China Publicity Chairs Liang Lin Luis Herranz Cees Snoek Shin’ichi Satoh Zi Huang Sun Yat-sen University, China Computer Vision Center, Spain University of Amsterdam and Qualcomm Research, The Netherlands National Institute of Informatics, Japan The University of Queensland, Australia Sponsorship Chairs Wangmeng Zuo Luhong Liang Ke Lv Harbin Institute of Technology, China ASTRI, Hong Kong, SAR China University of Chinese Academy of Sciences, China Registration Chairs Guorong Li Shuyuan Zhu Wenbin Yin University of Chinese Academy of Sciences, China University of Electronic Science and Technology of China Harbin Institute of Technology, China Finance Chairs Weiqing Min Wenbin Che ICT, Chinese Academy of Sciences, China Harbin Institute of Technology, China Contents – Part I Best Paper Candidate Deep Graph Laplacian Hashing for Image Retrieval Jiancong Ge, Xueliang Liu, Richang Hong, Jie Shao, and Meng Wang Deep Video Dehazing Wenqi Ren and Xiaochun Cao 14 Image Tagging by Joint Deep Visual-Semantic Propagation Yuexin Ma, Xinge Zhu, Yujing Sun, and Bingzheng Yan 25 Exploiting Time and Frequency Diversities for High-Quality Linear Video Transmission: A MCast Framework Chaofan He, Huiying Wang, Yang Hu, Yan Chen, and Houqiang Li Light Field Image Compression with Sub-apertures Reordering and Adaptive Reconstruction Chuanmin Jia, Yekang Yang, Xinfeng Zhang, Shiqi Wang, Shanshe Wang, and Siwei Ma 36 47 Video Coding Fast QTBT Partition Algorithm for JVET Intra Coding Based on CNN Zhipeng Jin, Ping An, and Liquan Shen A Novel Saliency Based Bit Allocation and RDO for HEVC Jiajun Xu, Qiang Peng, Bing Wang, Changbin Li, and Xiao Wu Light Field Image Compression Scheme Based on MVD Coding Standard Xinpeng Huang, Ping An, Liquan Shen, and Kai Li A Real-Time Multi-view AVS2 Decoder on Mobile Phone Yingfan Zhang, Zhenan Lin, Weilun Feng, Jun Sun, and Zongming Guo 59 70 79 89 Monaural Perfect Harmonic Musical Sound Separation Using Modified-CAM 905 The organization of the paper is as follows In Sect 2, we introduce a sinusoidal model of harmonic instruments and the CAM assumption Section presents our Modified-CAM form and our solution of overlapping partials in the perfect harmonic case (except the unison) in detail Section shows the experimental results of our system Finally, conclusions are drawn in Sect Sinusoidal Modeling and CAM Assumption 2.1 Sinusoidal Modeling A musical sound is composed of a series of periodic compound sounds It can be described as the summation of partials with different frequencies, amplitudes and phases Generally speaking, almost all pitched instruments’ harmonics are multiples of their pitches Within a period where sine waves are assumed constant, the sinusoidal model of an instrument source can be written as xð t Þ ẳ H X ah cos2pfh t ỵ uh ị 1ị h¼1 where ah , fh , and uh are amplitude, frequency and phase, respectively of the h (th) partial of the instrument source H denotes the number of partials in the source 2.2 CAM Assumption CAM assumes that all partials of a same instrument source have similar amplitude envelopes [13] has shown that envelope similarity between the strongest partial and other partials reduces as other partial amplitudes decay [3] has also stated that for real instruments, as the partial index increases, envelope similarity between the partial and the pitch decreases, yet it is still highly relative to the neighboring partial Assuming the time frame t, the frequency band f , the amplitude Eðt; f Þ, envelope similarity between partial A and B can be calculated by the inner product of their amplitudes: K P E ðt k ; f A Þ Á E ðt k ; f B Þ k¼1 s SimilarityfA ; fB ị ẳ s K K P P E ðt k ; f A Þ E t k ; f B ị kẳ1 2ị kẳ1 SimilarityfA ; fB ị is between and 1, where means the partials are completely identical, which is the principle of CAM assumption, and means they are irrelevant The larger the SimilarityðfA ; fB Þ, the higher similarity among the partials 906 Y Gong et al Modified-CAM Form in the Perfect Harmonic Case (Except the Unison) 3.1 Limitations of CAM Assumption in the General Case Li and Woodruff [13] have a system dealing with the general case by using CAM assumption Each instrument source has some non-overlapping partials (including the pitch) For an overlapping partial, despite both the phase shift and the amplitude ratio to the pitch are unknown, the frequency and amplitude envelope can be inferred by the pitch, and the unknowns can be solved in a least-square framework We make a refinement of their idea, which can be described as follows: Mixk ¼ kx X i¼1   ky m Á a à  X n Á bnà k m k non non non non non cxnon x cos À Da c y cos Db ỵ yj j ik i jk i m n jẳ1 3ị where the m (th) partial of source x and the n (th) partial of source y have collision The two sources have kx and ky non-overlapping partials respectively mà is the index of the closest non-overlapping partial to xm , and nà is the index of the closest one to yn All non-overlapping partials of x and y are represented as xnon and ynon i j In time frame k, and nÁbnÃnà k À Dbnon each estimation of the phases of xm and yn is mÁammà à k À Danon i j respectively with shifts ÀDanon and ÀDbnon unknown, likewise, each estimation of the i j non non non non amplitudes is cxnon xnon ik and cyj yjk respectively with ratios cxi and cyj unknown Mix i is the mixture signal In a general case that all instrument sources’ pitches are different and not have multiple relationships with each other, for example, two sources x and y which pitches are 220 Hz and 330 Hz respectively, the partials’ relationships are shown in Table Both two sources have some non-overlapping partials, including the pitches All these partials can be used by Eq (3) in CAM assumption However, when some sources’ pitches have some multiple relationships, things are much more complicated In a perfect harmonic case within one’s pitch twice as the other’s shown in Table 1, for example, x and y which pitches are 440 Hz and 880 Hz respectively, the amount of overlapping frequency regions is half partials of x with all partials of y overlapped This time, Eq (3) cannot be used obviously due to no partial of y is available Therefore, new solutions need to be found 3.2 Treatment by Combining Two Overlapping Frequency Regions In the perfect octave condition, assuming the overlapping region of x2 and y1 is frequency band A, and the overlapping region of x4 and y2 is frequency band B In spite of no clean partial of y, it is still a knowledge that y2 has a similar amplitude envelope Monaural Perfect Harmonic Musical Sound Separation Using Modified-CAM 907 Table Two overlap examples of the general case and the perfect harmonic case respectively Frequency/Hz 220 330 440 660 880 990 1100 1320 1540 1650 1760 1980 2200 …… Partial index of x Partial index of y 440 660 880 1320 1760 1980 2200 2640 3080 3300 3520 3960 4400 …… 10 Frequency/Hz Partial index of x Partial index of y 10 to y1 Hence, we assume that the amplitude envelopes are identical, and the frequency of y1 is known with y2 twice as y1 In this way, a Modified-CAM form can be shown as the following simultaneous equations:  mA Á amÃA k non MixAk ẳ cos DaAi ỵ yAk cosbAk DbA ị mA iẳ1   kx X mB amÃB k non cBxnon xnon cos À Da MixBk ¼ þ yBk cosðbBk À DbB Þ ik Bi i mÃB iẳ1 kx X  cAxnon xnon ik i 4ị yBk ¼ cun yAk where the meaning of each scalar and vector is the same as it appears in the Sect 3.1 yAk and yBk can be eliminated by these formulas, and then we can obtain   MixAk À ðEci qcAi ỵ Esi qsAi ị cos bBk q1 ỵ sin bBk q2 ị iẳ1   kx P ẳ MixBk Fci qcBi ỵ Fsi qsBi ị cos bAk q3 ỵ sin bAk q4 ị kx P iẳ1 5ị 908 Y Gong et al where mA Áamà k 1 cAnon cos Danon C B xi Ai qc Eci Ai B A C B cAnon sin Danon C sin C B Esi C B xnon B C à mA xi Ai C C B qsAi C ¼ B B C B ik mB Áamà k C; @ B non C; @ Fci A ¼ B non A B qc B c cos Da @ non Bi xi Bi A B xik cos mà C B A @ B non Fsi qsBi mB Áamà k c sin Da xnon Bi i xnon sin mà B B1 ik cun cos DbB B cun sin DbB C C ¼B @ cos DbA A sin DbA xnon ik cos A mÃA mA Áamà k q1 B q2 C B C @ q3 A q4 ð6Þ Written by matrix representation as 1 q1 Ac1 As1 À Bc1 À Bs1 À E1 F1 C B C C B q2 C B BÁÁÁ C C B q3 C B B C B C B Ack Ask À Bck À Bsk À Ek Fk CB 0C C B q4 C ¼ B B B C C AB @ÁÁÁ @ QA A @ A AcK AsK À BcK À BsK À EK FK QB ð7Þ It can be regarded as an over-determined system of homogeneous linear equations, where   Eci cos bBk Esi cos bBk Eci sin bBk Esi sin bBk ¼ ; Fci cos bAk Fsi cos bAk Fci sin bAk0 Fsi sin1bAk MixAk cos bBk Ack     B Ask C B MixAk sin bBk C E1k Á Á Á Eik Á Á Á Ekx k Ek C B C ¼ ; B @ Bck A ¼ @ MixBk cos bAk A; F1k Á Á Á Fik Á Á Á Fkx k Fk Mix bAk Bsk Bk sin1 Q Q A1 B1 1 B C B C q1 qcAi q3 qcBi B C B C C C B B B q1 qsAi C B q3 qsBi C C B B C B C QBi C QAi ¼ @ ; QBi ¼ @ ; QA ¼ B QAi C; QB ¼ B C B A A q2 qcAi q4 qcBi B C B C A A @ @ q2 qsAi q4 qsBi QAkx QBkx Eik Fik   ð8Þ Considering the linear correlation among the items of QA and QB , and q23 ỵ q24 ¼ in addition In this case, the least-square framework is not applicable We construct a gradient-descent framework, and solve the partial derivative of each unknown scalar: Monaural Perfect Harmonic Musical Sound Separation Using Modified-CAM K P  kx P 909  Sk Ack Eci qcAi ỵ Esi qsAi ị cos bBk C B B kẳ1  i¼1  C C BP kx K P C B B C Sk Ask Eci qcAi ỵ Esi qsAi ị sin bBk @y C B kẳ1 iẳ1 @q1 B 13 C C B B C Àq qc B @y C B Bi C B @q2 C B C B C C BP B kx K P Àq qs B Bi C7 C B @y C B C S Bc q Bs q ỵ F B C k k k ik B @DbA C B @ q3 qcBi A5 C C B k¼1 B i¼1 C C B B C C B B q3 qsBi C C B B C C B B C C B B C C B C B @y C ¼ B C B B @qcAi C B C K C BP B C B @y C B Sk ðÀEci q1 cos bBk À Eci q2 sin bBk Þ C B @qs C B C B Ai C B k¼1 C B @y C B P K C B @qc C B C Sk ðÀEsi q1 cos bBk À Esi q2 sin bBk Þ B Bi C B C B @y C B k¼1 C C B K B C @qs B Bi C B P C A B @ S Fc q cos b ỵ Fc q sin b Þ k i i C Ak Ak C B C B k¼1 K C BP C B Sk Fsi q3 cos bAk ỵ Fsi q4 sin bAk Þ C B k¼1 A @ ð9Þ Sk ¼ Ack q1 ỵ Ask q2 Bck q3 Bsk q4 Ek QA ỵ Fk QB 10ị where All partial derivatives should be In the gradient-descent framework with the initial value 0, we iterate until the Euclidean distance between the next iteration and the previous is less than 10−3, then we obtain all unknown scalars The reconstruction of y1 and y2 can be done by subtractions The envelope similarity between y1 and y2 is calculated by Eq (2), and the residual sum of squares (RSS) is calculated by the quadratic sum of Sk According to the assumption, it can be inferred that the separation performance is well when we get high envelope similarity and low RSS is meanwhile, otherwise the performance is poor We estimate the pitch of y by the equivalent of corresponding overlapping partial of x and the average of overlapping frequency bands from y1 to y10 preliminarily Each result shows that the similarity is a bit low and the RSS is a little high moreover Therefore, it needs to re-estimate the pitch of y much more accurately to improve the similarity and the RSS 3.3 Re-estimation of the Pitch of Source y As is mentioned before, it is unable to get any partial frequency of y, due to all partials colliding with corresponding partials of x According to limited separation performance of experiments, it is able to estimate y1 by the first ten partials of y, regardless of the 910 Y Gong et al overlaps, in this way we obtain ten different estimations Use these ten results in the gradient-descent framework, and iterate a certain number of times for each one, such as ten, and calculate each RSS of the last iteration Then choose two estimations yi and yj , which having the lowest RSS, making several weighted means After that, choose one weighted mean having the lowest RSS, iterating until the convergence Finally, the pitch of y is estimated more reasonably with much higher similarity and much lower RSS moreover, and the Modified-CAM form for the perfect harmonic case is much more robust 3.4 Treatment of the Rest Overlapping Frequency Regions In the presented model of gradient-descend with the Modified-CAM form, we use two overlapping frequency regions each time for the perfect harmonic case, which is quite different from one overlapping frequency region each time for the general case However, for example, the typical but the hardest condition, the perfect octave, when the first ten partials are taken from each instrument source, it needs to deal with at least five overlapping frequency regions of the mixture This time how to handle the rest overlapping frequency regions after the first treatment is still a problem If the recovered y1 and the recovered y2 are considered as known, then use the least-square framework to recover y3 , y4 and y5 , the quadric error will be led into Thus, for the rest overlapping regions, a cascade mode for the treatment of every two neighboring regions is designed For five overlapping frequency regions, it should be four treatments in total Because of repeated calculations of y2 , y3 and y4 , we take the average of two reconstructions as their waveforms respectively Corresponding x4 , x6 and x8 are recovered in the same way Thus, y1 and y2 , y2 and y3 , y3 and y4 , and y4 and y5 are all in accordance with the higher similarity between the two neighboring partials than two remote partials in the CAM assumption [3] (Fig 1) Mixture Estimated Pitches Partial Labeling The Perfect Harmonic Case Pitch Reestimation Phase GradientDescend Estimation Estimated Envelope Non-overlapping Partials Partial Recovery Separated Sources Fig System diagram for the perfect harmonic case (except the unison) Experiments 4.1 Database and Samples We use a database from the University of Iowa musical instrument sample database [17], which includes various kinds of western instruments, such as the woodwinds, the brass, the strings and the percussion Considering the complexity of the percussion, in which most are unpitched, we discard all samples We choose samples from nine Monaural Perfect Harmonic Musical Sound Separation Using Modified-CAM 911 instruments including Flute, Oboe, Clarinet, Horn, Trumpet, Trombone, Violin, Viola, and Cello All samples are in the actual playing having shifts and fluctuations Then we make mixtures of every two instruments according to the perfect octave interval, in this way the performance of the system can be tested straightly For each instrument, we choose five representative scales: C, D, E, G, and A Each instrument has a range over two octaves In our implementation, we have 95 clean samples and 430 two-instrument octave mixtures 4.2 Results We use signal-to-noise ratio (SNR) for evaluation Results are shown in Table The first row is the ΔSNR by forcing Li’s system [13] with both the amplitude and the pitch of y estimated by the mean of overlapping bands from y6 to y10 The second row is the ΔSNR by baseline with the pitch of y estimated by equivalent of the pitch of x The third row is the ΔSNR by baseline with the pitch of y estimated by the mean of overlapping bands from y1 to y10 The last row is the overall ΔSNR by our proposed system with re-estimation stage of the pitch of y It is clear that the re-estimation stage is of great importance to our proposed system, which achieves 3.2 dB and 3.4 dB higher than the two baselines separately, and 5.8 dB higher than forcing Li’s CAM system [13] Moreover, partials higher than the first ten need to be taken from the mixture in the follow-up experiments Table SNR improvement Forced General CAM System 6.72 Baseline 9.12 Baseline 9.31 Proposed System of Modified-CAM with Re-estimation 12.50 dB dB dB dB Conclusion In this paper, a monaural musical sound separation system is proposed to recover overlapping partials of perfect harmonic sound, especially handles the typical but the hardest condition of the case, the perfect octave interval Our strategy is based on a Modified-CAM form in a gradient-descend framework using estimated pitches, which greatly relaxes the limitations of the CAM assumption Experimental results also show that our proposed system achieves high improvements than that of predecessors When each overlapping partial frequency is estimated accurately, the performance has significant SNR improvements When an overlapping partial frequency cannot be inferred due to none of non-overlapping partials of the source, it can still be refined from a subset of the mixture’s overlapping partials And the re-estimation stage will achieve a decreased deviation from the truth and lead to improve the performance 912 Y Gong et al Acknowledgment This research was supported in part by the 973 Program (Project No 2014CB347600), the NSFC (Grant No 61672304 and 61672285), and the NSF of Jiangsu Province (Grant BK 20140058, BK20170033 and BK20170856) References Virtanen, T.: Sound source separation in monaural music signals Ph.D dissertation, TUT (2006) Li, Y.: Monaural musical sound separation M.S dissertation, OSU (2008) Viste, H., Evangelista, G.: A method for separation of overlapping partials based on similarity of temporal envelopes in multichannel mixtures IEEE TASLP 14, 1051–1061 (2006) Mellinger, D.K.: Event formation and separation in musical sound Ph.D dissertation, Stanford (1991) Virtanen, T.: Monaural sound source separation by nonnegative matrix factorization with temporal continuity and sparseness criteria IEEE TASLP 15, 1066–1074 (2007) Abdallah, S.A., Plumbley, M.D.: Unsupervised analysis of polyphonic music by sparse coding IEEE TNN 17, 1066–1074 (2007) Casey, M.A., Westner, W.: Separation of mixed audio sources by independent subspace analasis In: Proceedings of ICMC, pp 154–161 (2000) Klapuri, A.: Multiple fundamental frequency estimation based on harmonicity and spectral smoothness IEEE TSAP 11, 804–816 (2003) Virtanen, T., Klapuri, A.: Separation of harmonic sounds using multipitch analysis and iterative parameter estimation In: IEEE WASPAA, pp 83–86 (2001) 10 Every, M.R., Szymanski, J.E.: Separation of synchronous pitched notes by spectral filtering of harmonics IEEE TASLP 14, 1845–1856 (2006) 11 Bay, M., Beauchamp, J.W.: Harmonic source separation using prestored spectra In: Rosca, J., Erdogmus, D., Príncipe, J.C., Haykin, S (eds.) ICA 2006 LNCS, vol 3889, pp 561–568 Springer, Heidelberg (2006) https://doi.org/10.1007/11679363_70 12 Bregman, A.S.: Auditory Scene Analysis MIT Press, Cambridge (1990) 10 13 Li, Y., Woodruff, J.: Monaural musical sound separation based on pitch and common amplitude modulation IEEE TASLP 17, 1361–1371 (2009) 14 Han, J., Pardo, B.: Reconstructing completely overlapped notes from musical mixtures In: ICASSP, pp 249–252 (2011) 15 Ercan, M.B.: Musical instrument source separation in unison and monaural mixtures M.S dissertation, Bilkent (2014) 16 Rasch, R.A.: Description of regular twelve-tone musical tunings JASA 73, 1023–1035 (1983) 17 The University of IOWA Musical Instrument Sample Database http://theremin.music uiowa.edu/ Author Index Abeo, Timothy Apasiba I-526 Adil, Khan II-787 Ai, Chunling I-390 Amrani, Moussa II-68 An, Ping I-59, I-79, I-673 Bai, Cong II-347, II-583, II-818 Bai, Sichen II-981 Bao, Nan I-883 Cai, Bolun I-149 Cai, Xufen II-960 Cao, Wenlong I-303 Cao, Xiaochun I-14 Cao, Zigang II-488 Chang, Kan II-306 Chen, Guangyao II-447 Chen, Hui-Hsia I-515, II-847 Chen, Jiahui II-233 Chen, Jun II-626 Chen, Junjie II-560 Chen, Junliang I-873 Chen, Lu II-890 Chen, Mei I-773 Chen, Rui I-380, II-827 Chen, Wei II-663 Chen, Yan I-36 Chen, Yi II-634 Chen, Yimin I-743 Chen, Yixin II-172 Chen, Yuanchun I-629 Chen, Zhenzhong I-641, II-438, II-447 Chen, Zhikui II-56 Chen, Zhineng I-346 Chen, Zhumin I-893 Cheng, Feng II-702 Cheng, Long I-873 Cheng, Xina I-280, II-508 Cho, Ikhwan II-928 Cui, Yuhao I-754 Cui, Yukun I-117 Da, Qingan II-880 Dai, Feng II-911 Dai, Qionghai I-609, II-911 Deguchi, Daisuke I-497 Deng, Ziwei II-508 Doman, Keisuke I-497 Dong, Jing I-411, I-534 Dong, Peilei I-505 Dong, Zhen II-150 Dongyang, Zhao II-787 Du, Dapeng II-467 Du, Shaomin II-161 Du, Songlin II-539 Duan, Huiyu I-662 Duan, Lijuan I-258, II-254 Fa, Lingling I-802 Fan, Dongrui II-991 Fan, Mengdi I-505 Fan, Xiaopeng II-68, II-388, II-939 Fan, Yezhao II-24 Fang, Lou II-787 Fang, Shanshan II-818 Fang, Yuchun I-205 Feng, Weilun I-89 Fu, Canmiao II-477 Fu, Peipei II-498 Fu, Xiaowei I-722 Fu, Yunsheng II-3 Gan, Tian I-893 Gang, Li II-702 Gao, Chunchang I-854 Gao, Feng II-192 Gao, Ge II-634 Gao, Guanglai II-616 Gao, Hongchao I-423 Gao, Kun II-108 Gao, Weiyi I-303 Gao, Wen I-370, I-598, II-827, II-939 Gao, Xinwei II-939 914 Author Index Gao, Yuanyuan I-117 Ge, Jiancong I-3 Geng, Wenhao II-890 Gong, Yukai I-903 Gou, Gaopeng II-488, II-498 Gu, Ke I-629 Guan, Jian I-411, I-834 Guan, Yong II-275 Guo, Chunrong II-327 Guo, Guanjun I-227 Guo, Jie II-596 Guo, Kailing I-149 Guo, Qiang I-117 Guo, Sha I-598 Guo, Xiaoqiang I-138 Guo, Zongming I-89, I-764, II-674 Han, Jingfei II-378 Han, Jizhong I-423 Han, Qilong II-880 Han, Yahong I-464, II-108 Hao, Li II-233 Hao, Lingyun II-756 Hao, Pengyi II-347 Hao, Xingwei I-336 He, Chaofan I-36 He, Fazhi I-128 He, Gang II-766 He, Qi I-844 He, Siying II-477 He, Ying II-517 Hirayama, Takatsugu I-497 Hong, Hande I-873 Hong, Richang I-3, I-545 Hong, Zheran II-900 Hou, Yating II-517, II-948 Hou, Yilin II-508 Hu, Hai-Miao I-117 Hu, Jiagao II-222 Hu, Menghan II-24 Hu, Ruimin II-172, II-202, II-663, II-960 Hu, Xi II-634 Hu, Yang I-36 Hu, Yangyang II-327 Hu, Yupeng II-713 Huang, Kai II-98, II-139 Huang, Lei I-564, II-34 Huang, Qian II-397 Huang, Xingsheng II-79 Huang, Xinpeng I-79, I-673 Huang, Yujiao II-347 Ide, Ichiro I-497 Ikenaga, Takeshi I-280, II-508, II-539 Ivan, Andre II-928 Ji, Juan I-487 Ji, Qiujia I-293 Jia, Chuanmin I-47 Jia, Huizhu I-380, II-827 Jia, Xiaoyi I-149 Jia, Yunde I-773, II-150 Jiang, Feng II-3, II-68, II-787 Jiang, Hao II-766 Jiang, Hongbo II-306 Jiang, Jianmin I-863 Jiang, Junrong II-798 Jiang, Siyu I-743 Jiang, Xiubao I-598 Jiang, Xuesong II-683 Jifara, Worku II-3 Jin, Xin I-609 Jin, Zhipeng I-59 Jing, Chenchen II-150 Jing, Ming II-713 Jing, Qi II-674 Kawanishi, Yasutomo I-497 Ke, Shanfa II-626 Ke, Wei I-792, II-233 Khan, Adil II-3 Kong, Yongqiang II-192 Kou, Liang II-880 Kwong, Sam I-863 Lai, Hui II-735, II-745 Lan, Xuguang I-96 Lan, Yinhe II-407 Lei, Shiyao I-293 Lei, Xiaoyu II-129, II-337 Leng, Jiaxu II-571 Leng, Yonglin II-56 Leong, Hon Wai II-457 Li, Baokui I-346 Li, Bo II-222 Li, Changbin I-70 Li, Chaoxi I-873 Author Index Li, Chunyou II-596 Li, Dongxiao I-359, I-588 Li, Fu II-776, II-970 Li, Fufeng II-327 Li, Ge I-138, I-505 Li, Guohui I-652 Li, Guozhi I-883 Li, Hanqian II-296 Li, Hongzhu I-453 Li, Houqiang I-36 Li, Jiafeng II-890 Li, Jianing I-588 Li, Jianmin I-477 Li, Jianqiang I-873 Li, Kai I-79 Li, Longxi II-488 Li, Mading I-764 Li, Ping II-798 Li, Qingli I-172 Li, Qingnan II-172 Li, Wei I-477, I-792 Li, Weihai II-45 Li, Xiangwei I-96 Li, Xiaoqiang II-358 Li, Xixi II-438 Li, Xue II-327, II-550 Li, Xueqing II-713 Li, Yan I-883 Li, Yaoxian II-756 Li, Yongqiang I-844 Li, Yunsong II-766 Li, Zechao I-215 Li, Zhen II-488, II-498 Li, Zhenzhen II-498 Liang, Fangfang II-254 Liang, Qi I-577 Liang, Qiancheng I-883 Liang, Xiaohui II-161 Liao, Jian II-428 Lim, Jongwoo II-928 Lin, Chuang I-325 Lin, Weisi I-629 Lin, Xiangkai I-238 Lin, Zhenan I-89 Liu, Bin I-733, II-859, II-900 Liu, Chun II-920 Liu, Deyang I-673 Liu, Di II-438 Liu, Dilin I-783, II-683 Liu, Dongqin I-423 Liu, Fan II-397 Liu, Hongying I-172 Liu, Jiamin I-194 Liu, Jiaying I-764 Liu, Jie I-444 Liu, Jing II-24, II-723 Liu, Jingshuang II-378 Liu, Jue II-517 Liu, Jun I-433 Liu, Keke II-286 Liu, Li I-609 Liu, Na II-467 Liu, Qiang II-244 Liu, Rui II-756 Liu, Shaohua I-423 Liu, Shaohui II-920 Liu, Si-Xing I-526 Liu, Tingxi II-129 Liu, Weijie II-859 Liu, Xianming II-645 Liu, Xueliang I-3, I-545 Liu, Yan II-34 Liu, Yazhou II-264, II-286 Liu, Yiming I-336 Liu, Yuehu I-194 Lou, Fang II-3 Lu, Hong II-327, II-529 Lu, Ke I-444 Lu, Mingjie II-654 Lu, Ning I-652 Lu, Qing I-184 Lu, Tao I-107 Lu, Tong I-433 Lu, Xiusheng I-247 Lu, Yao II-89, II-129, II-337 Luo, Bin I-564 Luo, Chengwen I-873 Luo, Jian-Hao II-807 Luo, Linyao II-529 Luo, Tiejian II-571 Luo, Yanfei II-529 Luo, Yimin I-400 Lv, Kai I-792 Lyu, Xiaopan II-457 Ma, Ma, Ma, Ma, Ma, He I-883 Lin I-863 Ran I-673 Siwei I-47 Wei II-254 915 916 Author Index Ma, Yuexin I-25 Manhando, Edwin II-347 Mao, Jiafa II-583 Miao, Jun I-258 Min, Xiongkuo I-662 Ming, Yuewei II-244 Murase, Hiroshi I-497 Ng, Ruisheng II-368 Palaiahnakote, Shivakumara I-433 Pan, Jingui II-596 Pan, Lanlan I-107 Park, In Kyu II-928 Pei, Mingtao I-773, II-150 Peng, Qiang I-70 Philippe, Magali I-497 Pranata, Sugiri II-286 Pu, Jian II-13 Pu, Nan II-316 Qi, Shuhan I-411, I-834 Qi, Ziyu I-883 Qian, Guozhong II-948 Qian, Li II-316 Qian, Yinlong I-534 Qie, Chunguang I-227 Qin, Zheng I-269, II-98, II-139 Qing, Laiyun I-258 Qiu, Shuo II-254 Qiu, Song I-172 Qu, Yanyun II-296 Qu, Ying II-275 Ren, Tongwei II-34, II-869 Ren, Wenqi I-14 Rong, Wenge II-378 Rui, Jianwu I-303 Rui, Ting II-869 See, John II-368 Seok, Hochang II-928 Shang, Yan I-423 Shao, Jie I-3 Shao, Zhenzhou II-275 Shaohui, Liu II-3, II-787 Shen, Jianhua II-693, II-838 Shen, Liquan I-59, I-79, I-673 Shen, Shengmei II-286 Shen, Xiang-Jun I-526 Sheng, Hao I-792, II-233 Shi, Haichao I-534 Shi, Junzheng II-498 Shi, Kang I-811 Shi, Sheng II-991 Shi, Xiaoxue II-129, II-337 Shi, Zhiping II-275 Shu, Xiangbo I-802, I-903 Song, Dingyan II-634 Song, Mofei I-619 Song, Xuemeng I-893 Song, Yan I-802 Song, Youcheng I-619 Song, Yucheng II-663 Sori, Worku J II-787 Su, Yuanqi I-194 Sun, Chentian II-388 Sun, Jianguo II-880 Sun, Jun I-89, II-674 Sun, Li I-172 Sun, Quansen II-264, II-286 Sun, Wei I-662, II-723 Sun, Xiaoshuai I-247, I-693, I-712, II-119 Sun, Yihan II-571 Sun, Yuanliang II-756 Sun, Yuchao I-703 Sun, Yujing I-25 Sun, Zhengxing I-515, I-619, II-222, II-847 Suo, Jinli II-911 Takimoto, Hiroki I-497 Tan, Jindong II-275 Tang, Jian II-869 Tang, Jinhui I-802, I-903 Tang, Shun-Lei II-182 Tang, Wenbo I-325, I-712 Tang, Wenyi I-733 Tang, Xuehua II-529 Tang, Zhenhua II-306 Tao, Qiuyan I-359 Tu, Weiping II-626, II-663, II-702 Wang, Wang, Wang, Wang, Wang, Anran II-560 Bing I-70 Ci II-693, II-838 Daiyin I-161 Dong I-825 Author Index Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Wang, Feng I-577 Guolong I-269, II-98, II-139 Hanzi I-227, II-428 Haoqian I-754 Hong II-13 Huiying I-36 Jia II-24 Jiaming I-107 Jinshan II-626, II-702 Jinzhuo II-418 Kang I-96 Lianghao I-359, I-588 Mao II-244 Meng I-3 Qing I-258 Ran II-202 Renjie II-920 Ronggang I-138, I-238, I-370, I-505, I-598, I-683 Wang, Shanshe I-47 Wang, Shengke II-192 Wang, Shiqi I-47, II-645 Wang, Shu I-400 Wang, Shuang I-515, II-847 Wang, Shufeng I-293 Wang, Wei I-534 Wang, Weiqiang I-444, I-453, I-811 Wang, Wenhai I-433 Wang, Wenmin I-138, I-505, II-418 Wang, WenShan II-880 Wang, Wenwu I-411, I-834 Wang, Xi I-423 Wang, Xiaochen I-390, II-626, II-663, II-702 Wang, Xiaochuan II-161 Wang, Xingzheng I-754 Wang, Xu I-863, II-202 Wang, Xuan I-411, I-834 Wang, Xuehui II-911 Wang, Yanfang II-397 Wang, Yang II-388 Wang, Yasi II-119 Wang, Yiming II-378 Wang, Yuantian II-34 Wang, Yueming I-683 Wang, Zengfu I-184 Wang, Zheng II-766 Wang, Zhenyu I-598, I-683 Wang, Zhongyuan I-107, I-400 Wang, Zuchen I-205 Wei, Hongxi II-616 Wei, Xiu-Shen II-807 Wei, Zhaoqiang II-192 Wei, Zhengang II-192 Wen, Gao I-380 Wen, Zhenkun II-735, II-745, II-798 Weng, Zhenyu II-407 Williem II-928 Wong, Lai-Kuan II-368 Wu, Aming I-464 Wu, Chengyang I-883 Wu, Gangshan II-34, II-467 Wu, Gaoyu II-275 Wu, Hao II-847 Wu, Hao-Yi I-315 Wu, Huisi II-735, II-745, II-798 Wu, Jianxin II-606, II-807 Wu, Qianhao I-215 Wu, Si I-293 Wu, Song II-316 Wu, Tingzhao II-626 Wu, Wei II-981 Wu, Xiao I-70 Wu, Xinxiao I-703 Wu, Yafei I-390 Wu, Yanjun II-571 Wu, Yirui I-433 Wu, Yunjie I-619 Xia, Sifeng I-764 Xia, Yu I-693 Xiang, Lei II-457 Xiang, Xinguang II-212 Xiao, Gang II-583 Xiao, Guoqiang II-316 Xiao, Jing II-172, II-202 Xiao, Zhijiao II-79 Xie, Don II-827 Xie, Fanglu I-280 Xie, Xiaodong I-380 Xie, Zongxia I-834 Xing, Chang II-981 Xing, Shuo I-545 Xiong, Chang-Tai II-182 Xiong, Gang II-488, II-498 Xiong, Zhang I-792, II-378 Xiong, Zixiang I-107 Xu, Dan II-939 Xu, Feng II-517, II-948 Xu, Guisen I-683 917 918 Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Xu, Author Index Hui II-264 Jiajun I-70 Junfeng II-222 Kaiping I-269, II-98, II-139 Lisheng I-883 Long I-555 Longxiang II-859 Minjie II-327 Peng II-674 Shufang II-397 Wenyu II-970 Xiangmin I-149 Xiangyang II-467 Xin I-722 Xin-Shun I-315 Yi II-723 Yue I-359 Zhaochun I-444 Yan, Bingzheng I-25 Yan, Bo II-654 Yan, Jingwen II-172 Yan, Yan I-227, II-428 Yang, Cheng I-390, II-663 Yang, Fan II-827 Yang, Guogui II-970 Yang, Hanpei II-45 Yang, Kai II-3, II-68 Yang, Kewei I-515 Yang, Meng I-96 Yang, Min II-960 Yang, Ping II-818 Yang, Runkai II-991 Yang, Ruoyu I-487 Yang, Ruo-Yu II-182 Yang, Shaowu II-776 Yang, Wenhan I-764 Yang, Xiaokang I-662, II-723 Yang, Xuejun II-776 Yang, Yekang I-47 Yang, Yi I-754 Yang, Ying II-693 Yang, Yuanhang II-970 Yao, Hongxun I-247, I-325, I-693, I-712, I-783, I-825, I-844, II-119, II-683 Yao, Jian I-128 Yao, Wenbin I-161 Yao, Yunyu II-948 Ye, Feng II-397 Ye, Hao II-13 Ye, Long II-960 Ye, Shuxiong II-98, II-139 Ye, Xiaodan I-588 Yi, Xiaodong II-776 Yin, Jianping II-244 Yin, Lulu II-735, II-745 Yoon, Kuk-Jin II-928 You, Haihang II-991 Yu, Feiwu I-703 Yu, Jun I-184 Yu, Nenghai I-733, II-900 Yu, Wei I-325, I-712, I-783, II-119 Yu, Wennan I-703 Yu, Wenxin II-766 Yu, Xinguo II-457 Yuan, Haowei II-529 Yuan, Weihang II-847 Yuan, Xu II-56 Yuan, Yule II-477 Yunsheng, Fu II-787 Zeng, Qiang II-477 Zhai, Guangtao I-629, I-662, II-24, II-723 Zhai, Jian I-303 Zhang, Bin I-346 Zhang, Bo I-477 Zhang, Chen-Lin II-807 Zhang, Guoyin II-880 Zhang, Hong I-722 Zhang, Hongda I-117 Zhang, Hongyan II-838 Zhang, Hui II-616 Zhang, Jing II-890 Zhang, Junhui II-358 Zhang, Kao II-438 Zhang, Libo II-571 Zhang, Lifei I-802 Zhang, Liguo II-880 Zhang, Lin II-89 Zhang, Ming I-359, I-588 Zhang, Mingguo II-583 Zhang, Pan II-550 Zhang, Peng-Fei I-315 Zhang, Pingping I-863 Zhang, Qian I-773, II-306 Zhang, Sai II-869 Zhang, Shengdong I-128 Zhang, Shengping I-247 Zhang, Wanyi I-883 Zhang, Wenqiang II-327, II-529 Author Index Zhang, Xiaoyu I-534 Zhang, Xinfeng I-47, I-629, II-991 Zhang, Yan II-296 Zhang, Yanduo I-107 Zhang, Yang II-233 Zhang, Yanhao I-247, I-693 Zhang, Yi II-358 Zhang, Yihao II-418 Zhang, Yingfan I-89 Zhang, Yiwei I-138 Zhang, Yongdong II-911 Zhang, Yun I-863 Zhao, Chen I-370 Zhao, Debin II-388, II-645, II-920, II-939 Zhao, Dong I-555 Zhao, Dongyang II-68 Zhao, Hengying II-212 Zhao, Shuai I-346 Zhao, Wenbo II-645, II-920 Zhao, Ye I-545 Zhao, Yong II-477 Zhao, Yongping I-844 Zheng, JianWei II-818 Zheng, Jin I-227, II-428 Zheng, Nanning I-96 Zheng, Yanwei I-792 Zheng, Ying I-825 Zheng, Yingbin II-13 Zhong, Fangming II-56 Zhong, Hua II-56 Zhong, Jiang II-550 Zhong, Sheng-Hua II-34 Zhong, Sheng-hua II-79 Zhou, Hong-Yu II-606 Zhou, Jianshe I-444 Zhou, Jiantao I-629 Zhou, Jun I-722 Zhou, Mei I-172 Zhou, Tianfei II-89, II-337 Zhou, Yan I-641 Zhou, Yi II-981 Zhou, Yongsheng II-316 Zhou, Yuxuan II-756 Zhou, Zhong II-981 Zhu, Linan II-583 Zhu, Rong II-960 Zhu, Wenhan II-723 Zhu, Xiaotao II-397 Zhu, Xiaoyu I-893 Zhu, Xinge I-25 Zhu, Yucheng I-662, II-723 Zhu, Yuesheng I-161, II-407 Zhuo, Li II-890 Zou, Junhua II-869 919 ... Zheng-jun Zha Siwei Ma Chong-Wah Ngo Ruiqin Xiong Hefei Institute of Intelligent Machines, Chinese Academy of Sciences, China Peking University, China City University of Hong Kong, SAR China Peking... Polytechnic University, SAR China Fudan University, China ICT, Chinese Academy of Sciences, China Chinese Academy Sciences, China Harbin Institute of Technology, China VIII Organization Tutorial Chairs... Technology of China Harbin Institute of Technology, China Finance Chairs Weiqing Min Wenbin Che ICT, Chinese Academy of Sciences, China Harbin Institute of Technology, China Contents – Part I Best Paper