FUSIONOFGEOMETRICAL AND TEXTUREINFORMATIONFOR FACIALEXPRESSION RECOGNITION

A novel methodbasedongeometrical andtexture information isproposed forfacial expression recognition fromvideo sequences. TheDiscriminant Non-negative Matrix Factorization(DNMF)algorithm isapplied attheimageofthelast frameofthevideo sequence, corresponding tothegreatest intensity ofthefacial expression, thusextracting thetexture information. A Support Vector Machines (SVMs)system is usedfortheclassification ofthegeometrical information derived fromtracking theCandide grid overthevideo sequence. Thegeometrical information consists ofthedifferences ofthe nodecoordinates between theneutral (first) andthefully expressed facial expression (last) video frame. Thefusion of texture andgeometrical information obtained isperformed using SVMs.Theaccuracy achieved is98,7%whenrecognizing thesixbasic facial expressions. IndexTerms-Pattern classification, videosignal processing.