Parametric Reshaping of Human Bodies in Images

Project: Research

View graph of relations

Description

Image retouching aims to modify images in a proper way such that the resulting images are subtly more attractive. It has been widely used especially in the publishing industry. The commonly involved retouching tasks are often limited only to local modification of skins such as blemish and wrinkle removal. However, highly demanding retouching tasks like subtle reshaping of human bodies require global consistency of editing, which is rather challenging even for skilled artists. In fact, a professional artist has to use hundreds or even thousands of local editing operations to achieve a visually pleasing reshaping of a human body in single images. To make matters worse, the whole touchy, time-consuming process needs repeated if desired degree of retouching is required to change. The existing tools are already painfully awkward for reshaping human bodies in single images, let alone retouching human shapes in videos.This project aims for an easy-to-use image retouching system for photo-realistic reshaping of human bodies in images or videos. The key idea is that by effectively integrating a 3D whole-body morphable model into the reshaping process, we can easily enforce global consistency of editing by the 3D human model. More importantly, such integration allows us to parameterize the degree of reshaping via a small set of semantic properties, such as height and weight, relieving users of performing thousands of local editing operations. In this project, we plan to address various related issues and to build a complete body reshaping system. Our system is expected to allow both novices and professionals to easily achieve the desired degree of reshaping by simply moving several sliders and to produce professional editing results even for loosely-dressed humans with arbitrary poses in literally minutes.

Detail(s)

Project number7002533
Grant typeSRG
StatusFinished
Effective start/end date1/04/1031/07/12