Navigation

News

3 July 2020 

We are looking for two new doctoral coworkers (PhD students) for our lab. Please see advertisement of the positions here.

 

30 June 2020 

We have been granted access to large amounts of computing resources by the Northern German Network for HPC Computing (HLRN).

 

24 June 2020 

Our paper on learning of higher-order statistics for image encoding (Mousavi, Drefs, Lücke, 2020) has been accepted by the International Conference on Machine Learning, Optimization, and Data Science.

 

20 May 2020 
We have received top-up funding for the processing of SARS-CoV-2 EM microscopy images within our BMBF project SPAplus.

 

1 April 2020 
Our project "SPAplus" (collaborative BMBF project, 3 years) has started. We will investigate medical image processing using generative models.

 

25 March 2020 
Our paper "Phase transition for parameter learning of Hidden Markov Models" (Rau et al.) has been made available on arXiv.

 

15 March 2020 
Our research activities shift to home offices due to the Corona crisis

 

4 March 2020 
Our paper "Maximal Causes for Exponential Family Observables" (Mousavi et al.) has been made available on arXiv.

 

18 Feb 2020 
Our abstract "Optimal Inference of Sound Intensities and Sound Components Using Generative Representations" (Monk, Savin, Lücke) has been accepted for the ASA Conference in Chicago, where it will be presented as a talk.

 

6 Feb 2020 
Our project proposal "SPAplus" (collaborative BMBF project, 3 years) has been accepted and will be funded.

 

1 August 2019
Release of the open source software library "ProSper". The library contains a collection of algorithms for probabilistic sparse coding. For the source code see here. For a description see here.

 

1 July 2019
Our paper "k-Means as a Variational EM Approximation of Gaussian Mixture Models" has been published by Pattern Recognition Letters and is available online (PRLetters, arXiv).

 

2 April 2019
Our paper "k-Means as a Variational EM Approximation of Gaussian Mixture Models" was accepted for publication by Pattern Recognition Letters.

 

14-15 March 2019
Jörg Lücke gave a series of three lectures on Generative Machine Learning at the IK 2019 Spring School.

 

17 Jan 2019
Our paper "STRFs in primary auditory cortex emerge from masking-based statistics of natural sounds" (Sheikh et al.) has been published by PLOS Computational Biology.
 

16 July 2018
Our paper "Neural Simpletrons - Learning in the Limit of Few Labels with Directed Generative Networks" (Forster et al.) has been published by Neural Computation.
 

5 July 2018
Our paper "Truncated Variational Sampling for ‘Black Box’ Optimization of Generative Models" has been presented at the LVA/ICA 2018.

 

3 July 2018
Our paper "Optimal neural inference of stimulus intensities" (Monk et al.) has been published by Nature's Scientific Reports.
 

24 March 2018
Our paper "Evolutionary Expectation Maximization" (Guiraud et al.) has been accepted for GECCO 2018.
 

19 March 2018
Our paper "Truncated Variational Sampling for ‘Black Box’ Optimization of Generative Models" (Lücke et al.) has been accepted for LVA/ICA 2018.
 

5 March 2018
Our paper "Neural Simpletrons - Learning in the Limit of Few Labels with Directed Generative Networks" (Forster et al.) has been accepted by Neural Computation.
 

22 Dec 2017
Our paper "Can clustering scale sublinearly with its clusters?" (Forster & Lücke) has been accepted for AISTATS 2018.
 

30 June 2017
Our paper "Discrete Sparse Coding" (Exarchakis & Lücke) has been accepted by Neural Computation.
 

7 June 2017
Our paper "Models of acetylcholine and dopamine signals differentially improve neural representations" (Holca-Lamarre et al.) has been accepted by the journal Frontiers in Neuroscience.
 

25 May 2017
Our paper "Binary non-negative matrix deconvolution for audio dictionary learning" (Drgas et al.) has been accepted by the journal IEEE Transactions on Audio, Speech and Language Processing.

Contact

Head of lab

Prof. Dr. Jörg Lücke

+49 441 798 5486

+49 441 798-3902

W30 2-201

 

Secretary

tba

+49 441 798-

+49 441 798-3902

W30 2-202

 

Postal Address

Prof. Dr. Jörg Lücke
Arbeitsgruppe Machine Learning
Exzellenzcluster Hearing4all und
Department für Medizinische Physik und Akustik
Fakultät für Medizin und Gesundheitswissenschaften
Carl von Ossietzky Universität Oldenburg
D-26111 Oldenburg

Office Address

Room 201 (2nd floor)
Building W30 (NeSSy)  
Küpkersweg 74
26129 Oldenburg

Autonomous Cleaning of Corrupted Scanned Documents

Illustration of the generation Process (Click to enlarge)

We study the task of cleaning scanned text documents that are strongly corrupted by dirt such as manual line strokes, spilled ink etc. We aim at autonomously removing dirt from a single letter-size page based only on the information the page contains. Our approach, therefore, has to learn character representations without supervision and requires a mechanism to distinguish learned representations from irregular patterns. To learn character representations, we use a probabilistic generative model parameterizing pattern features, feature variances, the features' planar arrangements, and pattern frequencies. The latent variables of the model describe pattern class, pattern position, and the presence or absence of individual pattern features. The model parameters are optimized using a novel variational EM approximation. After learning, the parameters represent, independently of their absolute position, planar feature arrangements and their variances. A quality measure defined based on the learned representation then allows for an autonomous discrimination between regular character patterns and the irregular patterns making up the dirt. The irregular patterns can thus be removed to clean the document. For a full Latin alphabet we found that a single page does not contain sufficiently many character examples. However, even if heavily corrupted by dirt, we show that a page containing a lower number of character types can efficiently and autonomously be cleaned solely based on the structural regularity of the characters it contains. In different examples using characters from different alphabets, we demonstrate generality of the approach and discuss its implications for future developments.

Further Reading

  • Zhenwen Dai and Jörg Lücke (2014).
    Autonomous Document Cleaning – A Generative Approach to Reconstruct Strongly Corrupted Scanned Texts.
    IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(10): 1950-1962. (online accessbibtex)

  • Zhenwen Dai and Jörg Lücke (2012)
    Autonomous Cleaning of Corrupted Scanned Documents - A Generative Modeling Approach (pdf, bibtex)
    Proc. IEEE Computer Vision and Pattern Recognition (CVPR), 3338-3345, oral presentation.
    (highest CVPR 2012 reviewer score and Google Student Travel Award)

Copyright notice

The papers listed above have been published after peer review in different journals. These journals remain the only definitive repository of the content. Copyright and all rights therein are usually retained by the respective publishers. These materials may not be copied or reposted without their explicit permission. Use for scholarly purposes only.

Webhmlmastey3v2rhvy/d (petrghaa.ktnwiztydlts@d7muo/03hgl.dnv2eurux) (Changed: 2020-07-08)