CNN Based Costume Categorization Approach to Bootstrap the Costume Recommendations

Clothing products have increased significantly in terms of varieties, styles, colours, etc, with the continuing growth of e-commerce and the number of clothing products is growing. How to pick your favourite clothes from many items and to dress correctly and appropriately has become an issue that people must take into account in their everyday lives. This results in the creation of a user-oriented recommendation system, which addresses the problem of overloading information. In this research, we propose a costume recommendation system based on deep learning. In order to achieve this major objective, as the initial stage, in this paper, we present our CNN-based costume categorization approach to bootstrap the costume recommendations. In the first stage, the images were collected from five customers with different costumes. We collected around 500 photographs for the analysis. Then, we applied image-processing techniques to feed the images for the CNN model. Here, we applied different filters such as a median filter to reduce the noise. The median filter eliminated the noises without blurring the image. Finally, the CNN model was applied to train the dataset. We received a weighted average of 0.764, 0.696, and 0.705 for precision, recall, and F1 respectively.