Costume Colors Recommendation System Based On Skin-Color Recognition

With the rapid development of the economy,users increasingly pay attention to the personalization of costumes.Current costume colors recommendation systems usually recommend costume colors based on subjective judgments of users on their skin-color.On the other hand,the recommendation method,based on personal preferences and buying habits of users,lacks the professional matching guidance about costume colors.With the image recognition technology and deep learning developing rapidly,costume colors recommendation based on skin-color recognition has become a development trend in the field of costume sales.The costume colors recommendation system based on skin-color recognition designed in this paper employs image recognition technology to classify human skin-color and recommend costume colors based on the classification results.Firstly,the skin-color recognition model based on Unet-FC~4-SVM was designed to reduce the impact of illumination brightness and light-source color on skin-color recognition.Secondly,the Four Seasons Color Theory is employed to recommend costume colors to users with specific skin-color.Aiming at the problem that the illumination brightness and light-source color of the shooting environment affect the skin-color recognition in face pictures,this paper firstly designs the Unet-FC~4 illumination processing model.The advantage of the Unet algorithm is that it stitches features of the same dimension to prevent information loss.The experimental results show that the brightness enhancement algorithm based on Unet is compared with the histogram equalization and retinex algorithm.In terms of image quality evaluation indexes,such as MSE,PSNR,SSIM,its effect of brightness enhancement is better than the latter two algorithms;on the other hand,because the FC~4 algorithm employs a selection mechanism to avoid color-blocks with not obvious semantic from affecting the light-source color estimation,the light-source color removal algorithm based on FC~4,compared with the white balance algorithm,is also superior to the latter in the above three image quality evaluation indexes.Applying the skin-color classification criteria in the Four Seasons Color Theory,this paper divides the face skin-color into four types of spring,summer,autumn and winter,and designs a multi-classifier skin-color recognition model based on SVM.The model firstly employs a face detection algorithm based on Haar features and an Ensemble of Regression Trees(ERT)method to segment the pure skin-color region of the face to acquire useful information and achieve the purpose of data enhancement and reduce interference.Secondly extracts fusion features of skin-color and trains the multi-classifier based on SVM.Consequently,combined with the aforementioned Unet-FC~4 illumination processing model,a skin-color recognition model based on Unet-FC~4-SVM is obtained.The experimental results show that comparing the Unet-SVM and FC~4-SVM models,the skin-color recognition model based on Unet-FC~4-SVM has the highest accuracy.In this paper,the costume colors recommendation system based on skin-color recognition is programmed.Among them,the algorithm module(light processing,skin-color recognition)employs the Python language in the TensorFlow deep learning framework to realize the skin-color recognition of four seasons.In the costume colors recommendation module,the entity-relationship diagram of the users and the costumes table is designed based on the MySQL database,and the object storage system is employed to store files of costume pictures.Consequently,the user interaction,the algorithm module management,and the costume colors recommendation are implemented based on the Django web development framework.The costume colors recommendation system based on skin-color recognition designed in this paper can provide users with objective skin-color recognition results and give professional suggestions of costume colors.