Download Project (238 KB)
In this fast-track and uncertain world, the difference we can make with 3D view experience of someone is noteworthy and our intention is to draw them closer to the reality. Though people are attracted to the novelty, 3D is more traditional because it can more accurately reflects our everyday experience of interacting with the world. Our vision is to create a 3D view with lesser manual labor and have high standards of detailing in the output. 3D view consist of two principal components i.e. mesh and texture. The manual labor for the creation of each of them is cumbersome and time-consuming. Hence , we propose an automated and efficient technique to create the mesh and texture of the person from the input images which can be viewed in blender as a 3D model. Both of the components can be extracted from the a deep learning neural network which will train to analyze the given dataset and learn to predict a new texture and mesh combination for the desired output. Dataset contains of all the images from google with front and profile view. These images are annotated, mesh is generated manually with the help of blender software. Now, for the implementation part, the texture representation is generated with OpenCV library in python for Haar Cascades classifiers. The classification enables to get the exact face and eye boundary from the image. The mesh is detected with similarity detection from existing dataset and the most similar detection gives the mesh model. This process makes it faster to integrate and efficient for all types of devices.
Mehdi R. Zargham
Primary Advisor's Department
Stander Symposium project, College of Arts and Sciences
"3D face construction from front and profile 2D image" (2021). Stander Symposium Projects. 2154.
Presented April 22, 2021