Research On Real-Time Virtual Makeup And Recommendation Method Based On Image Processing

With the improvement of social civilization,women's requirements for make-up are getting higher and higher,virtual makeup technology has been used more and more widely.In this paper,we used the technology of image processing to realize the real-time makeup effect of video stream and realize the recommended function of makeup program.It realizes real-time interaction of human-computer when women purchase cosmetics.It also enhances the purchase speed and experience.The main research work of this paper is as follows:In the face detection stage,we combined the Haar-like feature with AdaBoost cascade classifier to detect face,and optimize the performance of the classifier in face deflection to improve the accuracy of detection by amplifying the positive samples with a certain deflection angle.In the phase of face alignment,the SIFT algorithm with better robustness was used to extract the local features.The feature points were aligned with SDM iterative refinement,and a more accurate feature point alignment model compared with the traditional AAM method was trained.In this paper,a virtual makeup method based on feature points and CIELAB color space layer decomposition is proposed and introduced from two phases.The rough makeup stage adopts contour extraction based on feature points to extract the contour.This method has better robustness than the method based on skin color model.Facial dermabrasion use triangulation method to remove non-processing area to keep local texture and speed up the processing.Fine makeup stage is based on CIELAB color space layer decomposition method,retaining the texture layer at the same time,the color layer used alpha channel fusion method to make the makeup effect more natural.In makeup recommended stage,we extract deep features based on DeepID network and chose the optimal network parameters by the experimental comparison,so the feature extraction capabilities is enhanced.Using the joint Bayesian model to calculate the similarity is more accurate than the softmax layer classifier that comes with the neural network.Finally,it retrieves the makeup image with the highest similarity in the beautiful image database as a recommended solution to provide a scientific and reasonable reference for the virtual makeup part.Experiments show that the virtual makeup technology adopted in this paper meets the user's requirements and can show a satisfactory makeup effect.