Abstract
Lucas-Kanade and active appearance models are among the most commonly used methods for image alignment and facial fitting, respectively. They both utilize nonlinear gradient descent, which is usually applied on intensity values. In this paper, we propose the employment of highly descriptive, densely sampled image features for both problems. We show that the strategy of warping the multichannel dense feature image at each iteration is more beneficial than extracting features after warping the intensity image at each iteration. Motivated by this observation, we demonstrate robust and accurate alignment and fitting performance using a variety of powerful feature descriptors. Especially with the employment of histograms of oriented gradient and scale-invariant feature transform features, our method significantly outperforms the current state-of-the-art results on in-the-wild databases.
Highlights
D UE to their importance in Computer Vision and HumanComputer Interaction, the problems of face alignment and fitting have accumulated great research effort during the past decades
The presented features can be separated in two categories: (1) the ones that are computed in a pixel-based fashion (e.g. Edge Structure (ES), Image Gradient Orientation kernel (IGO)), and (2) the ones that are computed in a window-based mode, they depend on the values of a larger spatial neighbourhood for each location (e.g. Histograms of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), Local Binary Patterns (LBP))
The results indicate that HOG-AIC and SIFT-AIC significantly outperform Discriminative Response Map Fitting (DRMF) and are more accurate than Supervised Descent Method (SDM)
Summary
D UE to their importance in Computer Vision and HumanComputer Interaction, the problems of face alignment and fitting have accumulated great research effort during the past decades. The static template image in the case of LK and a model texture instance in the case of AAMs. Since IC is a gradient descent optimization technique, the registration result is sensitive to initialization and to appearance variation (illumination, object appearance variation, occlusion etc.) exposed in the input and the target images [7]. We adopt the concept of highly-descriptive, densely-sampled features within the IC optimization and utilize multi-channel warping at each iteration of the IC optimization which does not greatly increase the computational complexity but significantly improves the fitting performance and robustness. We show that the combination of (1) non-linear least-squares optimization with (2) robust features (e.g. HOG/SIFT) and (3) generative models can achieve excellent performance for the task of face alignment.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
