International Joint Conference on Artificial Intelligence 2017

Fast Preprocessing for Robust Face Sketch Synthesis

Yibing Song1     Jiawei Zhang1     Linchao Bao2     Qingxiong Yang3    


Photo

MRF

MWF

SSD

Photo

MRF

MWF

SSD

Artist

Ours+MRF

Ours+MWF

Ours+SSD

Topic Background


In the data-driven framework for face sketch synthesis, the main issue is how to establish the accurate correspondence between training and input photos. Due to the lighting and pose variations, the inaccurate correspondence will bring in noticeable artifacts on the synthesized results.



Spotlight


Prior art methods : Correspondence establishment via feature extraction and matching.
Proposed : Formulation of online image segmentation into the offline alpha matting. Interactive adjustment of training and input photos.

Contributions: Our method can be integrated into existing methods to improve their robustness with ignorable computational cost.


Demos






Downloads


[Paper.pdf] : The paper.
[Results.zip] : Results on the benchmark datasets.
[Slides.pptx] : The slides.
[Poster.pdf] : The poster.


BibTex (DOI)


@inproceedings{song-ijcai17-faceSketch,
  author    = {Song, Yibing and Zhang, Jiawei and Bao, Linchao and Yang, Qingxiong},
  title     = {Fast Preprocessing for Robust Face Sketch Synthesis},
  booktitle = {International Joint Conference on Artificial Intelligence},
  pages     = {4530--4536},
  year      = {2017},  
}