Special Issue description

IEEE Transactions on Pattern Analysis and Machine Intelligence Special Issue:
Image and Video Inpainting and Denoising


Inpainting  is  the  field  of  research  devoted  to  the  development  of  methods  to  replace  missing  or corrupted  parts  of  the  image  data,  where  often,  only  small  regions  of  the  image  are  affected.  This problem has been widely studied within Computer Vision and Pattern Recognition. The interest of this community mainly raises from the number of problems/tasks that would be benefited from solving this problem.  Most  notably,  an  idea  inpainting  method  could  get  rid  of  the  occlusion  problem,  which  is present in almost all real world applications of scene understanding.  The problem has been approached with  a  wide  diversity  of  techniques.  The  first  methods  that  thrived  at  reconstructing  a  texture  or  at repeating  pattern  either  by  using  the  structure  of  the  textures  [1,  2,  3]  or  by  using  patches  [4], 
sometimes  also  called  image  quilting.  As  in  other  subfields,  recent  advances  in  computer  vision, partially led by Convolutional Neural Networks (CNNs) [5] have shown that these models are capable of  learning  a  great  amount  of  semantic  information,  useful  in  tasks  such  as  classification  or segmentation. There have also been advances in reconstructing missing information and denoising by using  denoising  autoencoders  [6].  The  work  of  [6]  is  capable  of  extracting  relevant  features  and reconstruct missing information based on them. More recently there have been considerable advances in generative models with Generative Adversarial Networks (GANs) [7]. For the problem of semantic inpainting this does not only provide yields a good generative model, but also a way to qualify some 
qualities of the results that traditional measures were not able to measure accurately. 

The recent and successful efforts for dealing with the inpainting and denoising problems put us in a perfect time to capture a snapshot of the state of the art at the early years of the deep learning era. With deep-learning-based solutions dominating this task,  with the recent release of associated benchmarks and with the organization of relevant workshops in top venues, a special issue in the topic would be an invaluable compilation that will pave the way for the development of the area in the forthcoming years. Accordingly,  we  are  proposing  a  special  issue  in  Transactions  on  Pattern  Analysis  and  Machine Intelligence devoted  to  cover    all  aspects  of  image  and  video  inpainting  and  denoising.   The  special issue  aims  to  compile  articles  providing  an  historical  review  of  the  field,  contributions  reporting  the 
latest advances in this research topic, and articles providing a perspective on the open  problems and challenges associated to image inpainting and denoising.  The special issue is associated to a couple of competitions  and  their  respective  workshops  collocated  with  WCCI  2018  and  ECCV  2018.  The  workshops scopes are totally aligned with the topic of the proposed special issue (see below); besides, the associated competitions deal  with the problems of pose estimation in occluded images, video de-captioning,   and   inpainting   and   denoising   for   fingerprint   verification.   We expect a crowded participation in both, challenges and workshops. The authors of best papers presented at the workshops and top ranking participants in the challenges will be invited to submit extended/improved versions of their work. Besides  with  the  open  Call  for  Papers  for  this  special  issue  on  all  aspects  of  image  and video inpainting and denoising we expect to define the current referent works in the field as well as to push its state of the art. 


Topics of Interest 
Advances in all aspects of computer vision and pattern recognition devoted to image and video inpainting. The proposed SI will focus on all aspects related to inpainting and denoising from videos and still images. More specifically, the committee encourages the submission of papers making fundamental or practical contributions to inpaining (or denoising) in connection with various computer vision topics including, but is not limited to:

  • Generative Adversarial Networks for image/video inpainting and denoising,
  • Pose estimation recovery from images/video with missing or occluded information,
  • Video de-captioning,
  • Inpainting/denoising for latent fingerprint recognition,
  • Unsupervised learning in image/video inpainting,
  • Future frame video prediction,
  • Structural image/video inpainting,
  • Textural image/video inpainting,
  • Combined structural and textural inpainting,
  • Multimodal image/video inpainting  (i.e., RGB-D)


[1]  Alexei  A  Efros  and  Thomas  K  Leung.  Texture  synthesis  by  non-parametric  sampling.  In: 
Computer  Vision,  1999.  The  Proceedings  of  the  Seventh  IEEE  International  Conference  on.  Vol.  2. 
IEEE. 1999, pp. 1033-1038.
[2] Alexei A Efros and William T Freeman. Image quilting for texture synthesis and transfer. In: 
Proceedings  of  the  28th  annual  conference  on  Computer  graphics  and  interactive  techniques.  ACM. 
2001, pp. 341-346
[3] David J Heeger and James R Bergen. Pyramid-based texture analysis synthesis. In: Proceedings 
of the 22nd annual conference on Computer graphics and interactive techniques. ACM. 1995, pp. 229-
[4] Lin Liang et al. Real-time texture synthesis by patch-based sampling. In: ACM Transactions on 
Graphics (ToG) 20.3 (2001), pp. 127-150.
[5]  Alex  Krizhevsky,  Ilya  Sutskever,  and  Georey  E  Hinton.  Imagenet  classication  with  deep 
convolutional neural networks. In: Advances in neural information processing systems. 2012, pp. 1097-
[6] Pascal Vincent et al. Stacked denoising autoencoders: Learning useful representations in a deep 
network  with  a  local  denoising  criterion".  In:  Journal  of  Machine  Learning  Research  11.Dec  (2010), 
pp. 3371-3408.
[7]  Ian  Goodfellow  et  al.  Generative  adversarial  nets.  In:  Advances  in  neural  information 
processing systems. 2014, pp. 2672-2680.


There are no news registered in