Related Experiment Video
Updated: Apr 8, 2026
11:34
High-resolution, High-speed, Three-dimensional Video Imaging with Digital Fringe Projection Techniques
Published on: December 3, 2013
16.2K
Transductive face sketch-photo synthesis
Summary
This study introduces a new transductive face sketch-photo synthesis method. It improves accuracy on test data by including it in the learning process, outperforming existing inductive methods.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Face sketch-photo synthesis is vital for law enforcement and digital entertainment.
- Current inductive learning methods show promise but struggle with high losses on test samples due to optimizing only training data.
Purpose of the Study:
- To present a novel transductive face sketch-photo synthesis method.
- To improve synthesis performance on test samples by incorporating them into the learning process.
Main Methods:
- A probabilistic model is defined to optimize reconstruction and synthesis fidelity.
- Alternating optimization is used for efficient model optimization.
- The method incorporates test samples directly into the learning framework.
Main Results:
- The proposed transductive method significantly reduces expected high losses for test samples.
- Improved synthesis performance on unseen test data was observed.
- Experimental results on the CUHK face sketch dataset validate the method's effectiveness.
Conclusions:
- Transductive learning offers a superior approach for face sketch-photo synthesis compared to inductive methods.
- The novel probabilistic model and optimization strategy enhance synthesis accuracy and reduce errors on test data.
