Neonatal Pain Expression Recognition Based On Gabor Wavelet And Sparse Representation

Recent medical research shows that newborns has a certain sense about pain, this earlypain will have some effect on the development and growth later. Since the neonatal does nothave the ability to express this pain feelings, facial expressions considered as a kind ofeffectively evaluation index to describe it. Therefore, developing a neonatal pain expressionautomatic evaluation system is meaningful and has broad application prospects. The mainlycompleted works are as follows:(1) First, I should focuses on Gabor wavelet transform as well as the different Gabor kernelwindow size on the influence of the recognition rate. Experimental results prove that use Gaborwavelet transform to extract feature can save the face image details commendably.(2) Based on the theory PCA, this paper uses the improved2D-PCA to reduct the dimensionof the Gabor feature value. Experimental results improve that compared with PCA,2D-PCA hasgreatly increased the operation efficiency.(3) Combined with the CS theory, take the sparse representation of the test image. Thispaper compared the advantages and disadvantages of different coefficient solve methods. Theexperiment proves that Newton method is best as well as the operation complexity is also thehighest.(4) At last compare Gabor+PCA with Gabor+2D-PCA extraction characteristics method.Experimental results show that2D-PCA dimensionality reduction method can greatly improvethe operation speed in the premise of does not affect the recognition rate. At the same time,compare Gabor+PCA with Gabor+down sample, the results showed that, the former methodrecognition can be up to94%-95%, far higher than the latter.Through the discussion of the algorithm and experimental results, this paper gets aconclusion as follows: use the Gabor+2D-PCA method extract characteristics, combined withNewton method to solve sparse coefficient solution can get a very good neonatal pain expressionrecognition rate.