Research On Facial Recognition Algorithms Based On Kinect

Facial recognition is a computer imaging technology that uses visual face feature information for identity authentication.There are many researches on facial recognition based on two-dimensional features,and many algorithms have been proposed.These algorithms have good performance in specific environment,but they cannot adapt to the increasingly complex demand needs.In recent years,with the advent of 3D scanning devices,three-dimensional depth information has been introduced to the field of facial recognition.However,due to the high costs of high-precision depth information acquisition equipment,facial recognition methods based on only three-dimensional information are difficult to popularize.In this thesis,after studying the existing two-dimensional face recognition algorithm and three-dimensional information,a face recognition method combining depth information is proposed.The method uses a small amount of depth information during the two-dimensional feature acquisition and processing.Thus,extracted features have good robustness,the algorithm runs more stable,simple,and efficient,and the recognition speed is improved.Firstly,the thesis makes a deep discussion on the process of facial recognition,including preprocessing,face detection,feature extraction,and difference measurement.It focuses on the application of principal component analysis method and local binary mode in the field of face recognition.With the help of Kinect equipment,the acquisition and expression of three-dimensional information were studied.On this basis,this thesis describes the proposed face recognition method combining depth information.The method quickly and accurately obtains the coordinates of the key points such as the point of nose and the root of nose on the face image by means of the depth information,and thereby separates eyes,noses and mouth image and calculates their distance,and separately extracts and classifies the features.The corresponding features are obtained by the image classifiers and the distance classifiers.Finally,the code is classified and identified as the whole person's face information.In order to verify the feasibility of the method in the actual environment,this thesis uses the Kinect device to collect the 2D-3D face image pair dataset for testing.The test results show that the proposed method has a significant improvement in the efficiency of the algorithm while ensuring the recognition accuracy.