Adam Novozámský

postdoc, doctor of natural sciences - mathematical egineering

Hello! I am currently a project assistant at the Computer Vision Lab (CVL) at TU Vienna and at the same time a postdoc at the Institute of Information Theory and Automation which administratively falls under the Czech Academy of Sciences. My research focuses on image analysis, medical imaging, segmentation, object detection, machine learning, and digital forensics.
I received my undergraduate degree in Computer Informatics in 2008, and a follow-up master’s program in Information Technology in 2010. In 2018 I defended my PhD in Computer Science and Applied Mathematics. All three degrees are from the Czech Technical University in Prague, Faculty of Nuclear Sciences and Physical Engineering.

download cv

Education : .

  • 2010-2018

    Department of Mathematics
    Faculty of Nuclear Science and Physical Engineering
    Czech Technical University in Prague

    Training workplace: Institute of Information Theory and Automation
    The Czech Academy of Sciences

    PhD degree in Applications of Natural Sciences - Mathematical Engineering
    Doctoral thesis: Selected Application Areas Of Image Processing: Image Forensics And Medical Imaging
    [ PhD thesis ]      [ presentation ]      [ co-authors statement ]

  • 2008-2010

    Department of Physical Electronics
    Faculty of Nuclear Science and Physical Engineering
    Czech Technical University in Prague

    Institute of Photonics and Electronics
    The Czech Academy of Sciences

    MSc degree in Applications of Natural Sciences - Engineering Informatics
    Master thesis: Software for Tomographic Reconstruction of Refractive Index Profile of Special Optical Fiber Preforms
    [ MSc thesis ]      [ presentation ]

  • 2003-2008

    Department of Physical Electronics
    Faculty of Nuclear Science and Physical Engineering
    Czech Technical University in Prague

    BSc degree in Applications of Natural Sciences - Engineering Informatics
    Bachelor thesis: The Electronic Guard of Physically Handicapped Persons
    [ BSc thesis ]

Research : .

Image Processing & Computer Vision

Out of all the five senses, our sight seems to be the most important. Sight and vision help people to connect with their surroundings.

Out of all the five senses, our sight seems to be the most important. Sight and vision help people to connect with their surroundings. People like to express themselves through pictures and approximately 65% of the general population are visual learners.
Nowadays, we live in the world with technologies such as Augmented Reality, Deep Fakes, or 3D-Medical Imaging. We meet images or videos everywhere around us, on the Internet, in newspapers and magazines, on television, at the doctor's office, or in private family albums.
No wonder that Image Processing (IP) and Computer Vision (CV) can be found in thousands of scientific, consumer, industrial, and artistic applications. This fantastic diversity of applications, modalities, and image types makes Image Processing such an exciting topic to study, where every project needs a genuine, innovative approach.
In my PhD study and on the postdoc position, I had an excellent opportunity to study many specialized topics in the fields of IP & CV.

Check projects I was involved with

Medical Imaging

Medical Imaging has a long history reaching back to the end of the 19th century, when in 1895 Wilhelm Conrad Roentgen took the first medical image, the radiograph of the left hand of his wife Bertha.

Medical Imaging has a long history reaching back to the end of the 19th century, when in 1895 Wilhelm Conrad Roentgen took the first medical image, the radiograph of the left hand of his wife Bertha. Only two months later, he performed the first clinical use and after half a year he presented a new kind of ray. His discovery had a profound impact on medicine. He was awarded the first Nobel Prize in Physics and many techniques were developed based on his approach.
Radiography, Computer Tomography (CT), Positron emission tomography (PET), Medical Ultrasound, and Magnetic resonance imaging (MRI), these five techniques are representatives of Medical Imaging. The last two are non-invasive and risk-free so far as we know. The first three are considered non-invasive as well, besides some risks associated with radiation exposure. Until recently, we have had to consider imaging in the visible spectrum as invasive, carried out through a surgery or classic endoscopy. But today's progress brings miniaturization and new technologies to overcome this issue.
Over the past several years, our department has worked closely with leading medical institutions in Czechia. Among other things, these long-lasting partnerships have resulted in two medical projects in which I was involved. Both are very specialized and deal with the display of the human body in the visible spectrum. The first one is the Wireless Capsule Endoscopy focused on imaging of the gastrointestinal tract. The second one is the Videokymography focused on imaging vocal cord function.

Check projects I was involved with

Image Forensics

Image Forensics is very closely related to the history of analog photography. One of the first tampered images originated a few decades after Niépce created the first photographs in 1826.

Image Forensics is very closely related to the history of analog photography. One of the first tampered images originated a few decades after Niépce created the first photographs in 1826. A nice example of the analog image splicing was created sometime around the year 1864, during the American Civil War.
Tampering with images is easy with today's software tools for Image Processing. A forger can add or remove important information which completely changes the message of a processed photo. The integrity of visual data is important for the credibility of news media and especially when used as an evidence in court or during criminal investigations. For this reason, we have observed a dynamic development in this research area in recent years. Over time, two branches of forensic analysis of digital images and videos have become established as essential. One of them is integrity verification (authenticity analysis), which determines if the material has been added, altered, deleted or changed in the image. The second one is image ballistics (source device verification), which identifies the source device of the acquisition process of the image.
Although past research in these areas has mainly focused on data hiding and digital watermarking approaches, today there is a growing alternative approach called the passive one, which does not need to embed any secondary data into the image.
Methods published in this research area have, for example, attempted to detect image splicing, traces of inconsistencies in color filter array interpolation, traces of geometric transformations, cloning, computer graphics generated photos, JPEG compression inconsistencies, file structure inconsistencies, etc. Typically, these methods are based on the fact that digital image editing brings specific detectable statistical or geometrical changes into the image. Others are looking for a distortion in the reality of the image scene.

The contribution of our department to this area of research is covered by these papers:
[Ministry of Interior : VG20102013064]Pizzaro: Forensic analysis and restoration of image and video data [H2020-EU.2.1.1. : 825227]IMD2020: A Large-Scale Annotated Dataset Tailored for Detecting Manipulated Images

Biological imaging

Image analysis methods and visualization are critical for understanding various features of cell and molecular biology.

Image analysis methods and visualization are critical for understanding various features of cell and molecular biology.
An increasing resolving power and efficiency of microscopic image acquisition hardware brings exponentially growing of biological image data sets. It poses great methodological challenges for image processing and quantitative analysis. Our department has a strong research program in this area of science and creates a technological background for other biological researchers from different angles of view.

I am deeply immersed
in Image Processing
for over 12 years and I still enjoy it because each problem has its individual solution.

There is always something new to discover.

Deep Learning for Computer Vision

Machine learning, especially Neural Networks architecture, is an extremely hot area in Artificial Intelligence and Computer vision.

Machine learning, especially Neural Networks architecture, is an extremely hot area in Artificial Intelligence and Computer vision. Over the last years, deep learning methods have been shown to outperform previous state-of-the-art techniques in several fields. On the other hand, a lot of scientists use Convolution Neural Networks without understanding their internal structure.
So one of my interests is understanding this internal structure and mechanisms of such machine learnig techniques and designing more efficient networks.

Publications : .

Automated Object Labeling For CNN-Based Image Segmentation

2020 IEEE International Conference on Image Processing (ICIP), p. 2036-2040

Deep learning-based methods for classification and segmentation require large training sets. Generating training data is often a tedious and expensive task. In industrial applications, such as automated visual inspection of products in an assemble line, objects for classification are well defined yet labeled data are difficult to obtain. To alleviate the problem of manual labeling, we propose to train a convolutional neural network with an automatically generated training set using a naive classifier with handcrafted features. We show that when the naive classifier has high precision then the trained network has both high precision and recall despite the low recall of the naive classifier. We demonstrate the proposed methodology on real scenario of detecting a car coolant tank. However, the proposed methodology facilitates collection of train data for a wider type of CNN based methods such as near-duplicate image detection or segmenting tampered areas of images.

[ PDF ]      [ BibTeX ]      [ Elsevier ]      [ ICIP2020 ]

IMD2020: A Large-Scale Annotated Dataset Tailored for Detecting Manipulated Images

2020 IEEE Winter Applications of Computer Vision Workshops (WACVW), p. 71-80

Witnessing impressive results of deep nets in a number of computer vision problems, the image forensic community has begun to utilize them in the challenging domain of detecting manipulated visual content. One of the obstacles to replicate the success of deep nets here is absence of diverse datasets tailored for training and testing of image forensic methods. Such datasets need to be designed to capture wide and complex types of systematic noise and intrinsic artifacts of images in order to avoid overfitting of learning methods to just a narrow set of camera types or types of manipulations. These artifacts are brought into visual content by various components of the image acquisition process as well as the manipulating process. In this paper, we introduce two novel datasets. First, we identified the majority of camera brands and models on the market, which resulted in 2,322 camera models. Then, we collected a dataset of 35,000 real images captured by these camera models. Moreover, we also created the same number of digitally manipulated images by using a large variety of core image manipulation methods as well we advanced ones such as GAN or Inpainting resulting in a dataset of 70,000 images. In addition to this dataset, we also created a dataset of 2,000 “real-life” (uncontrolled) manipulated images. They are made by unknown people and downloaded from Internet. The real versions of these images also have been found and are provided. We also manually created binary masks localizing the exact manipulated areas of these images. Both datasets are publicly available for the research community at http://staff.utia.cas.cz/novozada/db.

[ PDF ]      [ BibTeX ]      [ IEEE ]      [ WACV2020 ]      [ 1 citations ]

Detection of Copy-move Image Modification Using JPEG Compression Model

Forensic Science International vol.283, 1 (2018), p. 47-57

The so-called copy-move forgery, based on copying an object and pasting in another location of the same image, is a common way to manipulate image content. In this paper, we address the problem of copy-move forgery detection in JPEG images. The main problem with JPEG compression is that the same pixels, after moving to a different position and storing in the JPEG format, have different values. The majority of existing algorithms is based on matching pairs of similar patches, which generates many false matches. In many cases they cannot be eliminated by postprocessing, causing the failure of detection. To overcome this problem, we derive a JPEG-based constraint that any pair of patches must satisfy to be considered a valid candidate and propose an efficient algorithm to verify the constraint. The constraint can be integrated into most existing methods. Experiments show significant improvement of detection, especially for difficult cases, such as small objects, objects covered by textureless areas and repeated patterns.

[ PDF ]      [ BibTeX ]      [ Elsevier ]      [ 4 citations ]

Automatic blood detection in capsule endoscopy video

Journal of Biomedical Optics vol.21, 12 (2016), p. 1-8

We propose two automatic methods for detecting bleeding in wireless capsule endoscopy videos of the small intestine. The first one uses solely the color information, whereas the second one incorporates the assumptions about the blood spot shape and size. The original idea is namely the definition of a new color space that provides good separability of blood pixels and intestinal wall. Both methods can be applied either individually or their results can be fused together for the final decision. We evaluate their individual performance and various fusion rules on real data, manually annotated by an endoscopist.

[ PDF ]      [ BibTeX ]      [ SPIE ]      [ 9 citations ]

PIZZARO: Forensic analysis and restoration of image and video data

Forensic Science International vol.264, 1 (2016), p. 153-166

This paper introduces a set of methods for image and video forensic analysis. They were designed to help to assess image and video credibility and origin and to restore and increase image quality by diminishing unwanted blur, noise, and other possible artifacts. The motivation came from the best practices used in the criminal investigation utilizing images and/or videos. The determination of the image source, the verification of the image content, and image restoration were identified as the most important issues of which automation can facilitate criminalists work. Novel theoretical results complemented with existing approaches (LCD re-capture detection and denoising) were implemented in the PIZZARO software tool, which consists of the image processing functionality as well as of reporting and archiving functions to ensure the repeatability of image analysis procedures and thus fulfills formal aspects of the image/video analysis work. Comparison of new proposed methods with the state of the art approaches is shown. Real use cases are presented, which illustrate the functionality of the developed methods and demonstrate their applicability in different situations. The use cases as well as the method design were solved in tight cooperation of scientists from the Institute of Criminalistics, National Drug Headquarters of the Criminal Police and Investigation Service of the Police of the Czech Republic, and image processing experts from the Czech Academy of Sciences.

[ PDF ]      [ BibTeX ]      [ Elsevier ]      [ 6 citations ]

Teaching : .

Grants I worked on : .

Mobile diagnostic system for reduction of consumption and rational use of antibiotics for primary milk production

[2020 - 2022] Technology Agency of the Czech Republic : FW01010343

Objectives: The main objective of the project is to develop a dairy cow health control system, the ultimate goal of which would be to significantly reduce the use of antibiotics in the treatment and prevention of infectious mammary gland inflammation. Sub-goals: - Design a system of continuous microbiological diagnostics on dairy farms - Monitor the health and economic benefits of consistently applying the above system over a period of time. - Develop hardware equipment for the optical reading of color colonies of cultured microorganisms. - Develop a software solution associated with the reader to analyze the displayed microbial colonies. - Set up mutual links between the reader, its SW, the system administrator and the central database - Perform a thorough test of the whole system functionality.

project description

VKG 3.0

[2019 - 2021] Technology Agency of the Czech Republic : TH04010422

Objectives: The aim of the VKG 3.0 project is a new system for the diagnosis of vocal disorders, consisting of a new type of multi-line video camera and data processing software. The camera will allow to capture the vocal cords in a mode that detects their movement in several places at the same time, so the expert will have better idea of the vocal behavior and thus the ability to effectively make the correct diagnosis. The software for the proposed multi-line camera will be developed with the emphasis on data interpretation, special care will be devoted to the intuitive visualization of captured data. There are several features computed for each scan line. A large number of such data would not allow an effective evaluation of the finding.only the significant data will be fused.

project description

National Competence Center - Cybernetics and Artificial Intelligence

[2018 - 2020] Technology Agency of the Czech Republic : TN01000024

Objectives: The NCK KUI project aims to create a national platform for cybernetics and artificial intelligence which interlinks research and application oriented centers of robotics and cybernetics for Industry 4.0, Smart Cities, intelligent transport systems and cybersecurity. The connection of innovation leaders will raise effectivity of applied research in key areas, as advanced technology for globally competitive industry, ICT and transportation for the 21st century. NCK KUI is closely related to application sector and enables cross-domain collaboration, innovation development and technology transfer.

project description

PROVENANCE - PROviding VErificatioN AssistaNCE for New Content

[2018 - 2021] H2020-EU.2.1.1. : 825227

Objectives: PROVENANCE will develop an intermediary-free solution for digital content verification that gives greater control to users of social media and underpins the dynamics of social sharing in values of trust, openness, and fair participation. Specifically, PROVENANCE will use blockchain to record, in a secure and verifiable manner, multimedia content that is uploaded and registered by content creators or identified for registration by the PROVENANCE Social Network Monitor. The PROVENANCE Verification Layer will apply advanced tools for multimedia analytics (semantic uplift, image forensics, cascade analysis) to record any modifications to content assets and to identify similar pieces of content. A personalised Digital Companion will cater to the information needs of end-users. To help consumers navigate content and develop digital literacy competencies, an iconographic Verification Indicator will contextualise individual pieces of content with relevant information including when the content was registered, by whom, and any subsequent transactions. PROVENANCE will be co-created with diverse representatives of civil society across four distinct use-cases in the social media domain (citizen information seekers, citizen prosumers, factual content creators, and creative content creators). However, the findings will be applicable to any area in which social media and verification are important. The scientific and pragmatic insights gained through PROVENANCE will significantly advance the state of the art in intermediary-free solutions for content verification, understanding of information cascades and information sources on social media, the openness of algorithms, and user control over personal data. In so doing, it will lay the foundation for a new federated social network grounded in trust, openness, and fair participation. In addition, it will support the development of an observatory on information veracity and social media best practice under the ICT28 CSA.

project web page      project description

ASSISLT - Automated Software System In Speech-Language Therapy

[2018 - 2019] Technology Agency of the Czech Republic : TJ01000181

Objectives: The aim of our project is to create a software system to support speech therapy for adults and children with inborn and acquired motor speech disorders. The planned system focuses on individual treatment using exercises that improve tongue motion and thus articulation. The system will offer adjustable set of exercises recommended by a therapist, motivation by using augmented reality, evaluation of the performance of therapeutic movements, and session archivation. It will allow the therapist to evaluate the schedule and progress of the treatment. Linking the tongue movement and characters in a computer game will motivate children. The basic component of the system is the module evaluating the tongue motion based on image data from a commercially available camera.

project web page      project description

Automatic evaluation of videokymographic recordings for early diagnosis and prevention of vocal fold tumors

[2014 - 2017] Technology Agency of the Czech Republic : TA04010877

Objectives: The goal of the project is to develop a sophisticated software for videokymography (VKG) which will enable automatic evaluation of medical videokymographic recordings of vibrating vocal folds and arrive at correct medical diagnosis. Further goal is to develop a certified method of VKG evaluation to be used in clinical practice. VKG camera is an existing device which was developed in 1994 in collaboration of Czech and Dutch colleagues. It has been used for diagnosis of vocal fold vibration problems caused by various voice disorders, the most serious being vocal fold cancer. Currently, evaluation of VKG recordings can be done only by a highly experienced clinician. The evaluation is complicated and time-consuming. Therefore the method has not been widely used. Addition of a software for automatic evaluation will allow wider, more efficient and less time-demanding application of the method in clinical practice and an early and rather inexpensive diagnosis of tumorous states. It will allow detecting vocal fold cancer at an early stage when the treatment can be done noninvasively or by a simple surgical intervention so that the quality of life is preserved. The project is based on collaboration of four partners: clinical centre specialized in diagnosis and treatment of voice disorders (Voice Centre Prague, Medical Healthcom, Ltd), team of the inventor of the method of videokymography (dr. Svec, Department of Biophysics, Palacky University Olomouc), research institute specialized in digital image analysis (Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic) and a company experienced in diagnostic product development (STARMANS electronics).

project description

Capsule endoscopy in diagnostics of small bowel mucosal injury induced by nonsteroidal anti-inflammatory drugs

[2012 - 2015] Technology Agency of the Czech Republic : NT13532

Objectives: Prospective study is focused on identification of endoscopy, clinical and laboratory small bowel injury markers in long- term NSAID users. The patients with rheumatoid arthritis , osteoarthritis and healthy volunteers will be included into our study. The definition of the normal findings allows identification of the real NSAID induced injury. The detailed questionnaire concentrated on clinical signs will be filled with all participants. The laboratory tests and capsule endoscopy will be integral part of our study. The endoscopy findings will be scored according to the severity.

project description

Tools for imaging device identification, authentication, and image reconstruction

[2010 - 2013] Ministry of Interior : VG20102013064

Objectives: Software application, consisting of three modules for device identification, an authentication, and image reconstruction, respectively. The project output will enable an identification of the imaging device (digital cameras, camcoders), it will exclude the possibility of intentional post-processing changes of images or video. Finally, it will provide tools for quality improvement of analyzed image and video data using digital reconstruction methods.

project description

Development of methods for image analysis of photographs in digital and analog form for data authentication

[2010 - 2012] Ministry of Interior : VF20102012010

Objectives: Proposal of new software methods for image data authentication in still image record.

project description

Contact : .

info@adamnovozamsky.com skype: newcastlea +420-777-577-375
  • Department of Image Processing
  • Institute of Information Theory and Automation
  • The Czech Academy of Sciences
  • Pod Vodárenskou věží 4, Prague, Czechia
  • novozamsky@utia.cas.cz

  • Computer Vision Lab
  • Institute of Visual Computing & Human-Centered Technology
  • Faculty of Informatics
  • TU Vienna
  • Favoritenstr. 9/193-1, A-1040 Vienna, Austria
  • anovozamsky@cvl.tuwien.ac.at