Neonatal Pain Expression Recognition Based On Gabor Wavelet And Local Binary Pattern

Studies indicate that the repeated pains stimulations of the newborns can have some negativeimpacts on their future growth and development, what’s worse, it can cause permanent hurt oncentral nervous system. The traditional method, used to assess neonatal pain is conducted by healthprofessionals, but using the artificial methods, plenty of time and energy are spent to cultivateprofeddionals, which lead to less effective assessment, so there would be a wide range ofapplicationgs and potential market use to build a set of newborns facial expressions recognitionsystem.Feature extraction is the key technology in the procedure of neonatal pain recognition and itdetermines the recognition rate. This paper uses2D-Gabor wavelet changing and uniformity modeLBP algorithm to extract the feature of neonatal pain expression, then uses PCA method to reducethe dimensions of feature, and finally uses sparse representation as a classifier. Experiments showthat the proposed algorithm has a satisfied performance.The main work completed by this paper isas follows:（1）This paper has analysed and researched the method of feature extraction, has put forwardthe Feature Collecting Algorithm based on the combination of the2D-Gabor Wavelet and LocalBinary Pattern (LBP).（2）This paper has studied the impact on the recognition performance by the size of the corewindow（7*7，19*19，35*35，51*51，65*65） of the2D-Gabor Wavelet.The experiment resultsshow when the kernel window size is35*35recognition effect is the most ideal.（3）This paper has studied the influences of the division mode（3*3，4*4，5*3å'Œ6*5）ofuniformity mode LBP, the number of training samples and the number of trainingsampleson(300,400,500,600) on the recognition performance.（4）This paper has studied the influents of the different sparse coefficient solver on the results.Considering the overall effectiveness and recognition rate, the Truncated Newton Interior PointMethod adopted by this paper has better results.（5）This paper has studied the comparison of the recognition rate of four different featureextract methods under the same lab conditions and the test result indicates that the algorithm putforward by this paper can recognize the newborns pain facial expressions effectively and correctly,with the recognition rate reaching85.92%.