A Study on Pattern Recognition of Upper Body Clothes Using MediaPipePose

Abstract

In recent years, there are many studies that focuses on human region extraction and attribute recognition in images and videos. In clothing attribute recognition, there are many researches attempting to recognize the types and colors of clothing , as well as pattern recognition as one of the attributes in multi task classification. However, the accuracy of pattern recognition is still low. Nevertheless, pattern information is considered an important attribute since it is frequently used in clothing descriptions, as well as type and color. Therefore this study proposes a recognition system of the pattern of upper body clothing using deep learning. Our approach involves extracting the upper body clothing area from an image using MediaPipePose, followed by classification into five pattern categories utilizing a trained VGG16 model through transfer learning. We have performed the evaluation and confirmed that the average recognition rate achieved over 90%.