Search by image : clothing segmentation and retrieval
Date of Issue2019-04-23
School of Computer Science and Engineering
Our study aims to develop a product retrieval system for clothing items. More specifically, we formulate our task as a cross-domain content based image retrieval problem. Product search using traditional keyword matching can be inefficient and time- consuming due to the large semantic gap between low-level visual features such as shape and colors, and high-level intents of customers. Our system aims to bridge the gap between shoppers and retailers by offering a street-to-shop image retrieval system that supports searching by images. We divide our system into two main stages: 1) Segmentation: We segment the query image into individual clothing classes, and use them as regions of interest for subsequent retrieval. 2) Retrieval: For each of the clothing item, we use image retrieval techniques to fetch visually similar products. We show that, by training effective CNN models, our system achieves fast and accurate clothing retrieval result, and is able to handle challenging scenarios such as diverse viewpoints, occlusion, and complex background.
DRNTU::Engineering::Computer science and engineering
Final Year Project (FYP)
Nanyang Technological University