Robust and Accurate Shape Model Matching Using Random Forest Regression-Voting.

Area of Science:

• Computer Vision
• Machine Learning
• Medical Image Analysis

Background:

• Locating points on deformable objects typically involves fitting shape models to feature response images.
• Existing methods often rely on generative or discriminative models for pixel evaluation, which can be computationally intensive.

Purpose of the Study:

• To introduce and evaluate Random Forest regression-voting as a novel method for generating feature response images.
• To demonstrate the efficacy of this technique for fast and accurate shape model matching.

Main Methods:

• Utilized Random Forest regression-voting to generate feature response images for point localization.
• Applied the technique within the Constrained Local Model framework for shape model fitting.
• Compared performance against established alternative methods.

Main Results:

• Random Forest regression-voting efficiently produced high-quality response images.
• The method achieved fast and accurate shape model matching.
• Outperformed alternative techniques in hand joint annotation and facial feature point detection.

Conclusions:

• Random Forest regression-voting offers a superior approach for generating feature response images.
• This method achieves state-of-the-art accuracy in both hand joint annotation and facial feature point detection.