Face Detection & Replacement


And then there was Obama. The nefarious, Panoptic forces of the Internet had finally won, our lives categorized and remapped into submission unto his Holiness.

Histogram of Gradients (HOG) was applied for obtaining face descriptors, placing pixel gradients into orientation bins from 0-180° to generate a histogram. To learn face HOGs, SVM was used to train a model for classification of faces based on 6000+ faces and 170,000+ not-faces for training data.

Face replacement was done by detecting facial parts to obtain control points for Thin Plate Spline (TPS) morphing. Then, Poisson blending (third-party code) integrates the desired face into the image. My primary role was writing the HOG algorithm and implementing SVM.


CIS 581 Computer Vision & Computational Photography final project

Collaborator: Aayush Gupta