Research On The Algorithm Of Quality Detection Of Facial Skin Based On Computer Vision

Facial skin quality detection plays an important role in the beauty industry,nursing profession,dermatology and orthopedics.Facial skin quality detection includes skin diseases,wrinkles,acne,spots,and pore size.In general,The evaluation of different people for the same facial skin is generally different.However,it is very important to exclude subjective factors and provide objective results for people who specializes in skin surgeon.The traditional skin quality detection technology mainly relies on pixel color and threshold segmentation,which has several problems that low recognition rate,bad self adaptability and difficulty to be practice applied.To this end,in this paper,machine learning and deep learning technology are introduced into facial skin quality detection,improving the performance of detection.The key works and conclusions are listed as following:(1)A facial skin image classification algorithm based on Gabor filter and multi-features fusion is proposed.In the aspect of facial skin image recognition,illumination and noise are the two most important factors affecting the recognition rates.Aiming at this problem,this paper proposes to perform filtering on the skin samples by multi-scale and multi-directional Gabor filter,which can enhance the contrast between the detection target and the background,and reduce the effects of interference such as illumination and pores.Aiming at the problem that it is difficult for a single feature to reflect the complete information of skin image,this paper fuses local binary pattern histogram and directional gradient histogram to extract features,so as to effectively obtain the texture and contour of skin image.Experiments show that the recognition accuracy of fusion skin features is 93.8%after SVM classification,and the algorithm has been successfully applied to the Android framework of "Mirror of Beauty".(2)A facial skin image classification algorithm based on improved LeNet-5 convolutional neural network is proposed.Aiming at the problem that traditional machine learning can not extract the deep features of skin images,this paper proposes an improved LeNet-5 convolution neural network model based on a lot of research on the original LeNet-5.The improved LeNet5 network inherits receptive field size of the original LeNet-5,optimizes the network layers,increases the number of neurons,which can learn more features.Aiming at the problem of fewer skin samples,this paper effectively prevents the over-fitting problem of a picture by data enhancement,and further improves the accuracy of skin samples.Finally,the improved LeNet5 convolution neural network achieves 97.8%accuracy in skin image classification task,which is 7.6%higher than the original LeNet-5.(3)A system for detecting facial skin quality is built.Firstly,the system quickly finds the facial skin region through face detection,key feature points location and mask image construction.Then,using the skin classification model,the system can accurately identify facial"acne","wrinkles" and "color spots",which provides scientific data for workers in the field of skin.