Research Of Facial Feature Extraction In Video Based-on Environments Of Rain And Snow

Face recognition has always been a hot research of the field of pattern recognition andartificial intelligence. Facial feature extraction is the core steps of face recognition, and itsfeature extraction effect will directly affect the performances of the face recognition system.Outdoor camera systems are highly vulnerable to the impact of the rain and snow. And therain and snow will lead to a serious decline in the quality of the video image and impact facerecognition effects. Therefore, this article will do a research of "facial feature extraction invideo based-on rain and snow". The specific processes are as follows:First, we will do image preprocessing and performances analysis of the existingalgorithms of rain and snow removal. In order to reduce the amount of computation and thenoises, the video images are pretreated by graying, histogram equalization and medianfiltering. According to the optical and time performances of the rain and snow, the framedifference method and improved frame difference method will be applied for eliminating therain and snow. And we will analysis the results of simulation.Second, in order to solve the problems of the simulation results, an improved algorithmof eliminating the rain and snow in video images will be proposed. The improved algorithmadopts five consecutive frames of the video images to deal with the third one. We will first tocalculate the maximum and minimum gray value of the same pixel point in the five frames.And the pixel point which covered by rain and snow can be detected, when the difference ofthe maximum and minimum gray value is greater than zero. To calculate the distance betweenthe gray value of the same pixel point in the third frame and the maximum (minimum) grayvalue, respectively. When the distance between the minimum gray value is greater thanbetween the maximum gray value, to replace the value of pixel point in the third frame by theminimum gray value.Again, we detect and locate the faces and analysis the performance of the existing facialfeature extraction algorithms. We adopt the face detection and location method which basedon template matching and the edge weighted Hausdorff distance to detect and locate the facesin the video images. We will apply the PCA, ICA, LDA, KPCA and2DPCA facial featureextraction algorithm for face recognition, respectively. According to the existing problems ofthe simulation results, we will propose an improved facial feature extraction algorithm to solve them. The improved algorithm will define two matrices to store the training set, one is atwo-dimensional matrix and another is a three-dimensional matrix. Through performing the2DPCA algorithm on the two-dimensional matrix, we will get the feature space of2DPCA.We project the three-dimensional matrix onto the feature space and regarded the projectedresults as the training set of the KPCA algorithm, then performing KPCA. We project thestandardized testing set onto the2DPCA feature space and regarded the reshaped projecteddata as the testing set of KPCA algorithm. Also, we change the method of calculating thekernel function in the testing phase and performed the KPCA algorithm again. Finally, onenearest neighbor classifier based on Euclidean distance was used for face recognition.Finally, the improved algorithms of eliminating the rain and snow and the improvedfacial feature extraction algorithm will be applied on the face databases. Compared theirrecognition rates with the original recognition rate, respectively, to prove the effectiveness ofthe above two methods.