Fashion Model Recommendation System Based On Style Classification And Face Matching

In recent years,the main driving factor of the of Chinese economy development have been gradually shifting from export and investment to consumption.Now it presents a new consumption pattern: high quality,a combination of online and offline,individualization.As the main force of consumption,the apparel industry is developing rapidly,which makes the demand of garment model increase.With the development of the Internet,the network platform has become the main source of recruitment information.In order to increase their attractiveness and bring better user experience,many of the websites have launched personalized recommendation project.The personalized recommendation technology not only has the significance of academic research,but also has broader market demand.However,the recommendation techniques for fashion models in the garmend field did not shown up.In this paper,we research on the model recommendation filed.In order to design and implement the recommended system,we summed up the key factors of recruitment through communication with the fashion model recruiters.In this paper,we propose a fashion model recommendation system based on visual method.The system relies on the key factor of the recruitment.Firstly,the whole recommendation system will choose the sample picture as the reference,which uploaded by the recruiters.Then,the system filters the model pictures in database by the basic characteristics and style of the model.Finally,the system recommends the model through the sample or the fashion model’s face similarity which is decided by the individual needs of the recruiter.In the field of image processing,it is popular to research on the low-level features of images.However,the research on the image style classification which based on high-level semantic features is relatively rare.In this paper,through summarizing the characteristics of the garment model image,we combine the clothing style of the model with the color feature of the whole image which represents the emotion semantics to get the model style feature.Firstly,we carries on pose estimation to the garment model to get the sub graph of each part of the model and form the style of clothing through extracts the key features such as color,LBP,GIST and PHOG,integrated them with the whole image color feature to execute PCA dimensionality reduction.Finally,to get the model image classification results through the SVM training,which is execute the Genetic algorithm parameter optimization.In the part of face matching,after the face detection,feature point detection and normalization,we use a segmentation method which called ‘three chambers and five eyes’ to segment the face image.Then we extract the small blocks of image’s LBPH feature and combine all of these blocks with the whole image facial features,using the chi square distance to compute the face similarity.Finally,the system’s recommend result is decieded by style classification and face matching which is according to the personalized needs of the business.