Attila Fazekas: Face Detection and Facial Gesture Recognition

Some years ago the researches about multi-modal human-computer interfaces were a very hot area. Perhaps the intensity of research has decreased, but the results of many areas have already appeared at the technology level. Just think the face detection and facial gestures recognition. In my talk, I would like to give a short review of this area.

Ernst Schwartz: Deep learning in medical imaging

During the last decade, computer vision has been revolutionised by the application of deep neural networks. These methods have proven successful in diverse domains such as denoising, super-resolution, recognition, semantic segmentation and image registration to name a few. In this presentation, I will give a short overview of the major families of neural networks and their relevance for medical imaging problems. Based on recently published papers on discriminative and generative deep learning, I will present the specific challenges faced when applying neural-network based machine learning techniques to the medical domain, as well as the current approaches taken to tackle them.

Slavoljiub Mijovic: Fundamental Problems in Image Deblurring

The lecture is concerned with deconvolution methods for image reconstruction, knowing the mathematical model for blurring process during image formation. Special attention is devoted to ill-posedness of such problems and the ways how to fight it.

Tibor Lukic: Energy Minimization Models in Image Processing

Energy minimization models are often used in many image processing problems, such as tomography, image denoising or segmentation. The basic concept of this approach will be presented and discussed. Incorporating a priori information about the solution into the energy minimization model is called regularization. Special focus will be devoted to the development and analysis of possible regularization functions. Applications of regularized energy minimization models in discrete tomography will be presented.

Ivan Stajduhar: Everything you never wanted to know about machine learning, but were forced to find out

This is an introductory lecture to machine learning, tailored as a hands-on guide to its proper application. The lecture covers the following: data-driven modelling, model representation, optimisation, evaluation metrics, experimental setup, proper use of statistics, and improving model performance.

Debora Gil: Introduction to medical imaging processing I

Medical Imaging is one of the main application areas of computer vision and machine learning. In these talks we will give an introduction to medical imaging techniques with special focus on segmentation of anatomical structures in 3D volumes. During the 2 sessions we will introduce unsupervised methods for image object segmentation including preprocessing (image filtering), feature extraction (Gabor filters and difference of Gaussians), binarization (Otsu thresholding and k-means) and post-processing (morphological operations). We will also explain the main metrics for assessing the quality of segmentations.

Radu Ionescu: Towards Curriculum Learning from Images

Over the last few years, handcrafted models for object recognition from images have been replaced by trainable models based on deep learning, due to the impressive recognition performance of the latter models. Indeed, convolutional neural networks (CNN) achieve state-of-the-art performance for object recognition, semantic segmentation and related computer vision tasks. Convolutional neural networks are inspired by the hierarchical structure of the human visual cortex, but there are many aspects that are not captured by CNN. One of the most important aspects is that humans usually learn concepts on a progressive basis, starting with the easy concepts first. As we pass through the educational stages in school, we learn more and more advanced concepts that require our previously gained knowledge for proper understanding. Although CNN models reach very high accuracy levels for object recognition, the examples are usually presented in a random order during training. The main goal of this presentation is to showcase state-of-the-art CNN and GAN models trained on a curriculum learning paradigm, in which examples are presented gradually, from the easy ones to most difficult ones. In our recent work [Ionescu et al., CVPR16], we showed that the difficulty level of an image (with respect to a visual search task) can be automatically predicted. We show that predicted difficulty scores can be used to sort a set of images according to their difficulty, and use this information to train new CNN/GAN models in a curriculum learning setting, aiming to improve the recognition accuracy/generative capacity and the training time of state-of-the-art models.

Pedro Real: Parallel Topological Processing of Biomedical Images

Topology is the mathematics of relations, the mathematics of connectivity. In this talk, we address the problem of building an appropriate framework for designing (intrinsic) parallel algorithms to calculate advanced topological information (at level of representation or determination of features) of regions-of-interest (ROIs) of 2D and 3D biomedical images. Under the paradigm “Big Data within Huge Bitopological Scenario”, polynomial sequential algorithms are transformed into logarithmic parallel computational processes. We focus here on parallel algorithms for connected-component labeling and for determining topological representations of digital images like the classical Adjacency Tree (for binary images) or the Region-Adjacency-Graph (for color images). Finally, we also discuss possible topologically consistent versions of fractality and scale-space theories.

Vasile Gui: Robust Estimation Techniques in Image Processing

Modern computer vision systems require methods to extract useful information from visual data, like locations, shapes or projective information from different views of ascene. Often, the information of interest can be described by a model with parameters needing to be extracted from the unstructured image data. In this process the useful data is intermingled with irrelevant data, so effective solutions to separate the two are needed. The presentation reviews some useful robust estimation methods used in computer vision and related applications.

Mustafa Karhan: Image and Video Processing on Raspberry Pi Platform

It is aimed to learn the hardware and software components of Raspberry Pi platform, RaspiCam and USB camera connecting components of Raspberry Pi, operating systems for Raspberry Pi, basic Linux commands for image processing on Raspberry Pi platform, image and video acqusition and adjustment on Raspberry Pi, Image and video processing using MATLAB and Python on Raspberry Pi, example coding of basic image processing techniques which include filtering, feature extraction, morphological operations etc..., example implementations of image and video processing on Raspberry Pi platform.

By the end of this course, listeners will build image and video processing applicaitons that include image filtering, image enhancement, image segmentation, feature extraction. Example implementations will be mentioned ("Contact Angle Meter" and "Dielectric Phenomena Analysis using Image Processing Techniques on Raspberry Pi").

Carles Sanchez: Introduction to medical imaging processing II

Medical Imaging is one of the main application areas of computer vision and machine learning. In these talks we will give an introduction to medical imaging techniques with special focus on segmentation of anatomical structures in 3D volumes. During the 2 sessions we will introduce unsupervised methods for image object segmentation including preprocessing (image filtering), feature extraction (Gabor filters and difference of Gaussians), binarization (Otsu thresholding and k-means) and post-processing (morphological operations). We will also explain the main metrics for assessing the quality of segmentations.

Kalman Palagyi: Shape Description Using Skeleton-like Feature

Skeleton-like shape features (i.e., centerlines, medial surfaces, and topological kernels) are frequently used region-based shape features which summarize the general form of objects and represent their topological structures.

They play important role in various applications in image processing, pattern recognition, and visualization.

I shall define skeletons and present their properties.

Then the three major skeletonization techniques will be presented.

Finally some applications will be outlined.

Antal Nagy: Segmenting Head & Neck MR images

Magnetic Resonance Imaging (MRI) is a non-ionizing radiation imaging technique, which gives a high level of information on tissue changing in human body. In my talk, I will present manly results on MRI organs at risk segmentation in Head & Neck region performed in a project with my colleagues. I will also show liver tumor segmentation and characterization achievements. Besides these outcomes, I will demonstrate the pre- and post-processing techniques, which were applied, during our work.

Csaba Beleznai: Concepts, Algorithms and Practical Applications in 2D and 3D Computer Vision

Applications of Computer Vision slowly make the step towards practical use under demanding real-world conditions. This talk presents considerations from scientific and practitioner's point of view describing the process how a solution for a given task can be accomplished. These aspects are illustrated by several application examples targeting some selected challenging vision problems such as text detection/recognition in cluttered environments, 3D vision-based characterization of crowd movement and computer vision for autonomous driving and scene understanding. The talk attempts to provide a look under the hood of these systems by detailing considerations for the algorithmic choice, describing employed algorithmic concepts, presenting relevant implementation details, putting special emphasis on the interplay between Matlab and C++ for a rapid development process and demonstrating numerous results of completed real-time 2D and 3D vision systems.

Petru Radu: Hybrid Deep Learning Topology for Image Classification

The lecture is focused on investigation methods of improving the explainability and applicability of deep neural networks by applying evolutionary algorithms to train the deep neural networks architectures employed in image classification. Various methods for optimizing the training procedure of a deep neural network via evolutionary algorithms are explored. The lecture describes the research work conducted in two stages: first, the traditional training of the deep neural net for classification task is performed by various evolutionary algorithms. The use of neuroevolutionary algorithms allows for training parts of the deep neural network. Second, the final part of the deep network is replaced by a traditional non-trainable classifier such as k-Nearest Neighbor (k-NN).

Darian Onchis: Generalized Hough transform reinforced with Machine Learning for prospective medical diagnosis

In this talk, we present the recent activities of the SIMT (Signal, Image and Machine Learning Team) @ FMI and we concentrate on enhancing the accuracy of medical diagnosis using rapidly processed ultrasound and radiographic images. As an example, will present a lively-automatic method based on the anisotropic generalized Hough transform for estimating the left ventricle wall in echocardiographic images. The application of the transform is guided by a Gabor-like filtering for the approximate delimitation of the region of interest without the need for computing further anatomical characteristics like e.g. the apex. The algorithm is applying directly a deformable template on the predetermined filtered region and therefore it is responsive and straightforward implementable. For accuracy considerations, we have employed a support vector machine classifier to determine the confidence level of the automated marking. The presentation is concluded with medical discussions.