Automatic Retail Product Image Enhancement and Background Removal

Abstract

Retailers need good-quality product images with clear background on their Web sites. Most of these product images captured have diverse backgrounds, posing a challenge to separate the foreground from the background along with the enhancement of the product image. Currently, most of these activities are done manually. Our study proposes a computer vision (CV)- and machine learning (ML)-based approach to separate foreground (FG) and background (BG) from retail product images and enhance them. This automated process of BG/FG extraction involves two steps. A neural network (NN) classifier to identify if the BG has a monocolor gradient or not, followed by the separation of FG from BG and enhancement applied on the FG from the input image. Our results show 91% accuracy for BG/FG extraction and identifying the product region of interest (ROI).