000099671 001__ 99671
000099671 005__ 20190316233919.0
000099671 037__ $$aREP_WORK
000099671 245__ $$aBlind Audiovisual Source Separation Using Sparse Redundant Representations
000099671 269__ $$a2007
000099671 260__ $$c2007
000099671 336__ $$aReports
000099671 500__ $$aITS
000099671 520__ $$aIn this work, we present a method that jointly separates active audio and visual structures on a given mixture. This new concept, the Blind Audiovisual Source Separation (BAVSS), is achieved by exploiting the coherence existing between the recorded signal of a video camera and only one microphone. An efficient representation of audio and video sequences allows to build robust audiovisual relationships between temporally correlated structures of both modalities or, what turns to be the same, two parts of the same audiovisual event. First, video sources are localized and separated on the image sequence exploiting the temporal occurrence of audiovisual events and using a spatial clustering algorithm, without necessity of any previous assumption about the number of sources in the mixture. Second, the same audiovisual relationships together with a time-frequency probabilistic analysis allow the separation of the audio sources in the soundtrack, and, consequently, the complete Audiovisual Separation.
000099671 6531_ $$aAudiovisual processing
000099671 6531_ $$aBlind source separation
000099671 6531_ $$aSparse signal representation
000099671 6531_ $$aLTS2
000099671 700__ $$0242924$$g172395$$aLlagostera Casanovas, Anna
000099671 700__ $$0241005$$g150417$$aMonaci, Gianluca
000099671 700__ $$0240428$$g120906$$aVandergheynst, Pierre
000099671 8564_ $$uhttps://infoscience.epfl.ch/record/99671/files/TR-ITS-2007.pdf$$zn/a$$s3184285
000099671 909C0 $$xU10380$$0252392$$pLTS2
000099671 909CO $$ooai:infoscience.tind.io:99671$$qGLOBAL_SET$$pSTI$$preport
000099671 937__ $$aEPFL-REPORT-99671
000099671 973__ $$sPUBLISHED$$aEPFL
000099671 980__ $$aREPORT