Recognizing Smile Emotion Based On Fractional Fourier Transform And Local Binary Pattern

In recent years, with the great application of face expression recognition in the human-computer interaction, computer games and video conference, et al, expression recognition has gradually become the research focus. Besides, with the scientific rapid progress, especially the commercial use of smile detection cameras, smiling face as an important expression in the facial expression has received extensive concern, so how to improve recognizing smile emotion become serious problems. Fractional Fourier transform (FRFT) as the generalized form of the Fourier transform, have been widely used in signal detection, parameter estimation, phase recovery, image processing and so on. Most of the application in image processing research focuses on the digital watermark and encryption areas, while less involved in pattern recognition. Generally, face expression recognition system has three steps including face detection, feature extraction and feature classification, where feature extraction including feature dimensionality. The main work of this paper is verifying the effectiveness of the fractional Fourier transform as feature extraction in expression recognition. The specific research work in this paper is organized as follows:1. It introduces the main technical method of the expression recognition system, the basic definition and the characteristics of FRFT.Moreover, it analysis advantages of the existing discrete algorithm and lists practical dimension normalized algorithms which proposed by the Ozaktas and others. Because the research objects are two dimension images so we need to extend the one dimension of FRFT to two dimensions, which its relevant algorithm steps and properties are illustrated in detail.2. Recognizing smile emotion model based on the FRFT is constructed, recognizing smile emotion rate and total recognition rate are adopted to calculate simulation results. Due to the result of the FRFT transform is complex value information, so this paper lays emphasis on the research of the FRFT phase, amplitude and complex value information as a feature extraction method respectively; simultaneously we compare with Gabor filter feature which widely used and admitted in images recognition. Besides, it describes relevant theoretical knowledge and specific processing steps of the Adaboost classification algorithm. At last, the simulation results calculated on the same database do a comparative analysis to illustrate the efficiency of the method with low computing complexity.3. Due to high dimension of images, then feature dimension reduction is the necessary steps to improve the real-time of system, because three steps of expression recognition system has mutual influence on each other, so we need to use suitable dimension reduction method and classification criterion to adapt the method of feature extraction. In this paper the commonly dimension-reduction methods Discrete Cosine Transform and local binary pattern are used and simulated with using distance classifier and Adaboost classifier respectively to discuss the effectiveness of FRFT and LBP combined algorithm.