Extraction of Human Body Parts in Image Using Convolutional Neural Network and Attention Model

Abstract

In computer vision, human body parts extraction is a challenging task for many applications. In this work, we propose the algorithm to extract human body parts in images using the OpenPose system and attention model. The novelty of the proposed work is that algorithm is based on a convolutional neural network that uses a nonparametric representation to associate body parts with people in an image in combination with a new attention model that learns to focus on specific regions of different input features. The algorithm is a part of Smart Cropping system developed by us which aim is to cut the images and prepare e-commerce catalog.