Digital Image Processing
Snehasis Mukherjee
Face reconstruction|(3D) is a basic computer vision problem. Most systems dealing with this issue assume that there are plenty of facial images are available to them as an input and also, they have to address many challenges like large facial poses,expressions,alignment, etc. Generally these methods requires complex pipelines for model building and fitting. Our method address many of these limitations by training a Convolutional Neural Network (CNN) on an appropriate data-set consisting of 2D images and 3D facial models (dataset: AFLW). The CNN does not require accurate alignment and works with a single 2D facial image to reconstruct the whole 3D facial geometry by bypassing the construction and fitting of a 3D Morphable Model. This is achieved by a simple CNN architecture that performs regression of a volumetric representation of the 3D facial geometry from a single 2D image. This also demonstrates how facial landmark localization can be included and into the framework and help improve or enhance reconstruction quality for the case of facial expressions.
Click Here: Report