Related Experiment Video
Updated: Nov 16, 2025
11:34
High-resolution, High-speed, Three-dimensional Video Imaging with Digital Fringe Projection Techniques
Published on: December 3, 2013
15.9K
FASHE: A FrActal Based Strategy for Head Pose Estimation
Summary
FASHE, a novel approach using fractal-based methods, achieves state-of-the-art head pose estimation without deep learning. This method offers a robust alternative for accurately determining head rotations from single images.
Area of Science:
- Computer Vision
- Image Processing
- Machine Learning
Background:
- Head Pose Estimation (HPE) is crucial for various applications, including best-frame selection.
- Deep learning methods dominate HPE but require extensive training.
- Alternative, non-deep learning approaches may offer competitive performance.
Purpose of the Study:
- To introduce FASHE, a novel head pose estimation method.
- To demonstrate the efficacy of Partitioned Iterated Function Systems (PIFS) for HPE.
- To provide an accurate and robust HPE solution without extensive training.
Main Methods:
- FASHE utilizes PIFS to represent facial auto-similarities.
- A single frontal reference image is used to extract domain blocks.
- Pose estimation is performed by matching fractal codes using Hamming distance.
Main Results:
- FASHE surpasses state-of-the-art results on Biwi and Ponting'04 datasets.
- Performance approaches top methods on the challenging AFLW2000 database.
- Successful in-the-wild operation demonstrated on the GOTCHA Video Dataset.
Conclusions:
- FASHE offers a powerful, non-deep learning alternative for head pose estimation.
- The PIFS-based approach provides high accuracy and robustness.
- FASHE shows promise for real-world, in-the-wild applications.
