An Intelligent fashion recommendation system

Abstract

In the present as a result of the rapid rising in living standards, most people have been tempted to develop their interest in shopping. As a result of that, nowadays there is a huge demand for garments and the number of people who pursue fashion has increased. Since there are different types of garments available in the market people have to figure out what needs to buy and this will lead to trying garments repeatedly, which consumes more time for the selection. On the other hand, even though most of the sellers have online stores the real benefits of online stores cannot be obtained because the consumers always have a doubt whether the purchase will be matching with him or her. Besides all of these, it is somewhat impossible for the merchant to identify the real customer demand and create an outfit based on each person’s satisfaction. In the present most of the available recommendation systems will recommend the clothes for a user based on the activities or the behavior of the other users who used the system previously by considering that all users behave simultaneously. The current user’s personal preferences will not take into account. But when it comes to the cloth recommendation, it is very crucial to consider the users’ personal preference. This paper presented an automated way of recommending outfits based on the user image by directly incorporating the visual signals into the recommendation objective. This research provides more insights on how convolutional neural networks can be used for the feature extraction phase from fashion images and evaluate the output from different CNNs. To achieve better results than the available neural network this research is proposing a hybrid approach of using both ResNet and VGG. The final evaluation of the system proves that the hybrid approach is having a positive impact on achieving more accurate results than the existing systems.