Renderizado de coloración capilar utilizando inteligencia artificial

Abstract

The importance in the world of hair coloring has become a trend and people who decide to change their hair color go to an esthetics where they usually ask for recommendations without knowing the final result. The purpose of creating a hair color rendering application is to provide a digital tool that shows color recommendations through a process of hair colorimetry by previewing a hair tone virtually. The application collects three physical characteristics: eye, skin and hair color that are stored in a MySQL database. The Support Vector Machine algorithm uses the TensorFlow and Keras libraries for the hair colorimetry process to predict hair color. The DeepLabv3 model segments the hair of a loaded image resulting in a mask. Finally, a process that couples the results of the previous algorithms is applied to display the customized hair color using the RGB model. The evaluation of the hair colorimetry algorithm uses cross-validation and obtains a value of 0.78 in training and testing, being able to predict the color adequately. The quality metrics for the DeepLabv3 model present average results: recall 98%. Dice Coefficient 98% and IoU 97%, considering that the model has good ability to segment hair in images. The application offers benefits of personalized and realistic counseling, providing a favorable change in the user's appearance by combining theoretical and technological creativity.