Interactive System for Fashion Clothing Recommendation

Abstract

An interactive system for fashion clothing recommendation is upcoming and exploring area that deals with image retrieval. Basically it aims to have efficient online shopping systems that take clothes image as an input and automatically retrieve similar clothing images from massive collection of clothing image dataset. In addition, such systems are trained to generate relevant style tags or annotations for the query image. Existence of factors like heterogeneous style attributes, body poses and appearances, background, etc. generate several challenges for both the tasks. Past few years have identified part-based representations as useful tool in this domain. A survey of such systems is presented in this manuscript drawing attention towards their contribution. An attempt is also made to summarize must have features of fashion clothing recommendation systems thus presenting the future directions in this domain. Based on the inferences drawn, an image based fashion clothing retrieval system is designed as well as implemented. It works with two features (color histogram and attribute descriptor) and is divided into two main modules. One retrieves matching clothing pair and second allows user the flexibility to change a few style related attributes that helps in result refinement as per the user requirement. It effectively uses the concepts of neural networks and its applicability is validated using a sub-set of self-collected image dataset with 1,100 images from freely available DeepFashion dataset. The developed system is evaluated using visual examination of retrieved images and is observed to achieve satisfactory accuracy values.