Image Multi-Feature Fusion for Clothing Style Classification

E-commerce platforms are evolving rapidly, and consumers have adapted to shop online more effectively by utilizing product images on shopping websites. In the case of apparel, consumers pay attention not only to the information about practical attributes but also to the information about style attributes. In this study, to specifically analyze the respective style recognition capabilities of the depth feature extractor for studying different clothing styles, an empirical investigation was conducted on four commonly utilized networks: AlexNet, InceptionV3, ResNet50, and VGG16. To complete style classification, we propose the IMF model, which integrates shallow neural networks subsequent to feature engineering to incorporate the benefits of basic features and depth features. This process reinforces the basic features through secondary extraction and merges them with depth features to generate style features, ultimately accomplishing the objective of clothing style classification. Furthermore, a DeepCluster based unsupervised learning approach is used in this study as a comparison. Its classification outcomes are compared with those of the IMF model to authenticate the efficacy of IMF model in this study.