Detection And Analysis Of Facial Skin Based On Deep Learning

With the progress of the times,people pay more attention to their appearance,skin care has become a hot topic.In the detection and analysis of skin quality,the traditional method is that professional cosmetologists do analysis rely on their knowledge of nursing and long-term diagnostic experience.With the development of science and technology,many large skin care companies and beauty parlors began to study the facial skin quality tester,using machine instead of manual to detect facial skin condition.However,at present,the facial skin tester on the market is mainly based on small sensors and traditional digital image processing algorithm technology products.The emergence of in-depth learning has greatly promoted the development of two-dimensional image pixel-level classification in the field of computer vision.Inspired by this,this thesis takes the combination of deep learning method and digital image algorithm as the research focus,tries to segment the skin lesion features such as pigments,red blood filaments and so on,and carries out quantitative analysis of these skin problems.The main contents of this thesis are as follows:(1)Face image processing based on digital image algorithm.Before the convolution neural network,it is necessary to annotate the collected images and preprocess them to a certain extent.In the process of labeling,this thesis mainly uses two methods,labelme labeling toolkit and HotoShop software.In the process of preprocessing,many digital image processing technologies are used,such as color block extraction based on HSV color space,direction gradient histogram and local binary method to obtain image texture features.(2)Semantic segmentation of face image based on Deeplabv3+.After image preprocessing,it will be sent to the convolutional neural network for training and learning together with the annotated image.At this stage,four network structures with unique advantages in semantics segmentation are selected as the benchmark structure,namely Fully Convolutional Networks(FCN),U-net,DFN and Deeplabv3+.Through performance comparison,Deeplabv3 + is selected as the core segmentation algorithm.(3)Evaluation and analysis of facial skin image.Firstly,600 groups of facial skin images are selected to make data sets,and then compare and analyze two evaluation models,that is,the evaluation model based on data statistics and the classification method based on convolution neural network.Finally,the evaluation model based on data statistics is selected as the product evaluation model.The main skin tasks involved in this topic are: wrinkles,pore,superficial pigments,red blood silk,acne,deep pigments,chemical fluorescent agents and so on.(4)The whole platform of hardware and software is built.Data images are collected by a unique image acquisition device and transmitted to Android applications in customized tablets by Bluetooth communication.Then,image information is uploaded to the server by a tablet connected to WiFi network for algorithm processing.The processed images are returned to Android applications together with the analysis results and displayed to users through a visual interface.In addition,the application will also give specific nursing advice and recommend appropriate skin care products,as well as provide before and after the skin change comparison.This platform realizes the complete testing process of face skin quality detection from photo taking to result display.Through the combination of digital image processing and deep learning technology,the user's own skin condition,causes,solutions and prevention programs can be displayed to the maximum extent.