Research And Implementation Of Facial Skin Quality Evaluation Method Based On Machine Learning

A reasonable evaluation of the quality of facial skin plays an important role in the field of dermatological diagnosis,cosmetic evaluation and skin care and nursing.However,at present,the skin quality testers on the market both domestic and foreign have the disadvantages of being expensive,large and inconvenient to carry.The "Haier Magic Mirror" exhibited by Haier Company,one of the modules of wisdom and health,has a single point of view on skin quality evaluation.The "you look good today" App with leading role of "photo skin test",the evaluation of acne is only given the severity.Based on the algorithms of machine learning,this thesis evaluates the quality of skin quickly,quantitatively and objectively through facial images.The main research results are as follows:(1)A facial skin quality evaluation method is proposed.At present,the evaluation methods of facial skin quality are mostly different skin quality grades,and the evaluation results are not quantitatively analyzed.In this thesis,facial skin quality assessment methods,including two parts,they are wrinkle index which assigns different weights of different regions and acne index which is based on the area and the number of acne.This evaluation method can not only indicate the level of skin quality,but also by comparing the final evaluation results to further compare the skin quality differences with the same level.(1)Proposed a facial skin quality evaluation method.At present,the evaluation methods of facial skin quality are mostly different skin quality grades,and the evaluation results are not quantitatively analyzed.Facial skin quality includes many aspects such as skin diseases,wetness,gloss,pore size,pH,wrinkles and acne.Based on the users’ most concerned wrinkles and acne,a facial skin quality evaluation method is proposed in this thesis,including two parts,they are wrinkle index which assigns different weights of different regions and acne index which is based on the area and the number of acne.This evaluation method can not only indicate the level of skin quality,but also by comparing the final evaluation results to further compare the skin quality differences with the same level.(2)A wrinkle recognition algorithm based on LBPH and GLCM is proposed.Different people’s facial skin wrinkles with different deep,even if the same person’s different wrinkle areas are also different,so the wrinkle recognition difficulty is the feature extraction.Since the wrinkles belong to the detail information of the skin texture,and facial wrinkles tend to have the directionality,the local binary pattern histogram can effectively describe the image local features,while the gray level co-occurrence matrix can reflect the information about the direction and magnitude of the image.Therefore,combining the advantages of the two features,this thesis combines the local binary pattern histogram features and gray level co-occurrence matrix to extract the wrinkle feature.(3)An unsupervised acne recognition algorithm based on fluctuating eigenvalues is proposed.The existing acne recognition algorithm usually first performs morphological processing on a single image,selects a threshold and compares every pixel with a threshold to determine whether it is acne.Due to the comparison of single pixel value,the method is susceptible to noise.In this thesis,the unsupervised acne recognition algorithm based on fluctuating eigenvalues is proposed.Firstly,the image is divided into blocks and the eigenvalues of the current pixels are calculated according to the surrounding pixel values.The final classification threshold is based on a large number of sample statistics,this method can effectively eliminate the interference caused by noise in a single pixel or a single image.At the same time,the algorithm improves the real-time of acne recognition.(4)A random forest acne recognition method based on grid search algorithm is proposed.Random forest algorithm due to the random amount of random sampling process,it is not sensitive to noise data,effectively reducing the impact of noise on the recognition results,and achieved good results.In this thesis,the experimental results of the analysis using cross-validation,orthogonal experiments,analysis of variance and comparative analysis methods.The results show that the skin quality evaluation algorithm used in this thesis can effectively evaluate the skin quality and achieve the expected results.