Research On The Technology Of Rotating Face Expression Recognition

Facial Expression Recognition (FER) can be applied in many fields, such asvirtual reality, video conferencing, customer satisfaction studies in broadcasting andnetwork services, human-computer interaction, robotics control and drive statusdetection, and has very important applications value. It has increasingly become a hotresearch.In FER, facial expression images captured from different circumstances mayhave some rotation in plane compared with the reference image. The method to solvethis problem of the traditional FER system is correcting the expression images in theprocess of image preprocessing, to make the eyes at the same level, and then throughthe processes of the feature extraction and classification, the recognition result isobtained. This research deeply on this key problem, proposed a new kind of solutionwhich can avoid correcting the rotation angle of the image preprocessing, andimprove the recognition speed, without reducing the recognition effect. The mainresearch work and innovations are as follows:1. This paper describes the theoretical and framework structure of FER system,at the same time, summed up the latest research in FER field. The every step of facialexpression recognition is summarized, with emphasis on feature extraction algorithmis studied.2. A new FER system based on Polar Fourier Transform (PFT) is proposed, andthe system has two characteristics: the system don’t need the image correction inimage preprocessing; the feature extraction algorithm which can transform therotation into the translation, and then with the help of DFT the influence of translationcan be eliminated is not sensitive to the rotation, so that we can extract the rotationinvariant feature. The two characteristics of the system are verified by the theory, andthe simulation experiment is carried on Ryerson University RML database, the resultshows that the system has good robustness.3. The FER system is constructed based on the pseudo-polar Fourier transform(PPFT). The advantages of the PPFT and the value of replacing the PFT are prove by the theory. The PPFT is constructed by a pseudo polar network to approximate thePFT, and the PPFT has fast algorithm but the PFT don’t, so the PPFT can improve thespeed of calculation. At the same time, The PPFT can approach the PFT by two steps-horizontal interpolation and radial interpolation which can ensure the recognition rate.The relationship between recognition rates and the angle of rotation images isanalysis by experiment, and the experiment proves has more advantages.