Slovakia

This presentation shows the results of our research based on the comparison of approaches of automatic detection and localization of an optic disc and macula using on deep learning. Information about their position helps to diagnose diabetic retinopathy - a disease that causes visual impairment and leads to blindness. Detecting the symptoms of this disease manually is a demanding process, but thanks to advances in computer vision, it can be automated. Compared methods base on Faster R-CNN, YOLO and SSD detectors are used to detect the optic disc and macula in the eye fundus images. We test the methods on three publically available datasets of fundus images - “DRIVE“, “IOSTAR“ and “ODIR“. We created own dataset containing 500 fundus images as a combination of mentioned databases, where the ground truth position of the macula and optic disc were marked by an ophthalmologist.

IWSSIP 2021

Detection of optic disc and macula in retinal images using deep learning methods

optic disc

macula

diabetic retinopathy

object detection

deep learning

ICPR 2020 is the premier international conference on Pattern Recognition, gathering leading researchers from around the world to present their latest achievements in Pattern Analysis and Recognition, applied Artificial Intelligence and Machine Learning. The topic of the year is " Putting Artificial intelligence to work on Patterns". 
For the first time, ICPR has employed a two-round review process similar to journal submissions, with new papers allowed to be submitted in either the first or second round and papers submitted in the first round and not accepted allowed to be revised and re-submitted for second round review. 

In this edition of the **ICPR 2020** Conference, there are 1407 Oral/Poster presentations organized in five Tracks: 
- Artificial Intelligence, Machine Learning for Pattern Analysis; 
- Biometrics, Human Analysis and Behavior Understanding; Computer Vision, Robotics and Intelligent Systems; 
- Document and Media Analysis; and 
- Image and Signal Processing. 

The **ICPR 2020** Main Conference program also includes 7 Keynotes from prominent world-renowned scientists, 8 Tutorials, 28 Demos, 8 Challenges, 2 Panels, as well as an Industry/Sponsor Exhibition. In addition, 41 Workshops on a range of important current and emerging research topics complete the ICPR 2020 program. 


ICPR 2020

technical paper


IWSSIP is an international conference that brings together researchers and developers from both academia and industry reporting the latest scientific and theoretical advances to discuss and debate major issues and state-of-the-art aspects on systems, signals and image processing.

###### For the session discussion, each author will be seated at individual breakout tables.  Please find which table they will be at here:
###### [QA table assignments in Gather.town](https://drive.google.com/file/d/1hBHOD_KF7NSsaTcpYcHCZslXJvQVxJQe/view?usp=sharing)

"We create a compact 3D model-based gait feature out of
2D images using DensePose. We also design a CNN to show that the
obtained gait feature can be used to improve recognition. Results are
based on the publicly available CASIA-B dataset."

Gait Recognition with DensePose Energy Images

Monitoring plant growth with computer vision is an important topic in plant science. This monitoring can be challenging when plants are located in outdoor conditions due to light variations and other noises. On other hand, there is a lack of annotated datasets available for such outdoor environments to train machine learning algorithms while indoor similar datasets may be more easily available. In this communication, we investigate, for the first time to the best of our knowledge in plant imaging, how to take benefit from model trained in fully controlled environment to build model for an outdoor environment. This is illustrated with the prediction of seedling development from top view RGB images.  

Deep learning-based detection of seedling development from indoor to outdoor

"Due to the omnipresence of computers today, the need for forensic experts has increased. Much of the knowledge associated with digital forensics has become common and has found its way to universities through courses or study programs.
As with many other domains related to computer engineering, computer forensics is based on theoretical concepts which must become integrated into skills and those skills serve as foundations of competence. For example, computer forensics experts must be able to recover data from a deleted file or reconstruct the sequence of actions in a cyberattack. Assessment is, therefore, a necessary precondition of effective computer forensics training and outcomes evaluation. Skills related to computer forensics should be acquired through practice and assessed in an authentic environment, yet that is difficult to achieve since the development of a large number of custom problems for assessing them is timeconsuming. This paper describes an approach to addressing this issue with a system for automatic generation and scoring of individualized and authentic computer forensics problems. Problem types and system implementation details are described in the paper."

Computer forensics problem generator for competence assessment

"This work presents the computer vision and machine learn-
ing techniques used in an implementation of a prototype system to aid
on understanding DICOM exams. In such a way, all aspects included
from the image prepossessing to identification and final classification of
bone anomalies by the digital bone scintigraphy images are commended
in details. It is possible to state that these used techniques are very
promising, because although, there are few patients already included for
leaning, the results obtained in the tested phase achieved more than
seventy percent of correction when used with Decision Tree (DT) and
k-Nearest Neighbors (K-NN) classification algorithms."

Presenting a system to aid on the scintigraphy bone metastasis analysis using DICOM files

Sparse representation and compressive sensing theory is wide used in image processing with the core idea that given a transformation that describes a signal in sparse domain can be achieved a moore compact and thus compressed form of the signal through optimization. In this research discrete orthogonal moments transforms are applied in sparse representation and compressive sensing theory under the implementation of transformed-based compressive sensing. Finally a comparison take place between image moments transforms and DCT in compression and denoising. 

Moment transform-based Compressive Sensing in Image Processing

Particle colliders are machines built to probe fundamental questions in physics. The properties of the produced particles are measured by complex experiments which use a wide variety of devices, such as the calorimeter system. In the ATLAS experiment at LHC, the Tile Calorimeter (TileCal) comprises about 10,000 readout channels that amplify, shape, and sample each signal every 25 ns. Since LHC collisions occur every 25 ns, and due to the increase of the luminosity level, signals from adjacent collisions may be read out within the same TileCal readout window, deforming the expected signal, and degrading the energy estimation efficiency. Therefore, this work compares the performance of the currently available methods for TileCal energy estimation using LHC collision data. Different pile-up conditions are considered. The results show that the performance in terms of the uncertainty of the energy estimation can be improved up to 35% in high occupancy readout channels.

Energy Reconstruction Techniques in TileCal under High Pile-up Conditions

Automatic cardiac MRI segmentation, including left and rightventricular endocardium and epicardium, has an essential role in clinical  diagnosis  by  providing  crucial  information  about  cardiac  function. Determining heart chamber properties, such as volume or ejection fraction,  directly  relies  on  their  accurate  segmentation.  In  this  work,  we propose a new automatic method for the segmentation of myocardium, left, and right ventricles from MRI images. We introduce a new archi-tecture that incorporates SERes blocks into 3D U-net architecture (3DSERes-U-Net). The SERes blocks incorporate squeeze-and-excitation operations into residual learning. The adaptive feature recalibration ability of squeeze-and-excitation operations boosts the network’s representational power while feature reuse utilizes effective learning of the features,which  improves  segmentation  performance.  We  evaluate  the  proposedmethod on the testing dataset of the MICCAI Automated Cardiac Diag-nosis Challenge (ACDC) dataset and obtain highly comparable resultsto the state-of-the-art methods

Segmentation and Quantification of Bi-Ventricles and Myocardium Using 3D SERes-U-Net

This presentation evaluates results of an ongoing research employing novel offline end-to-end speech recognition models for Slovak language. Transition to ESPnet's Transformer neural network architecture is introduced along with a brief theoretical introduction to end-to-end based speech recognition.

Transition from hybrid to end-to-end speech recognition in Slovak utilizing Transformer neural network architecture

In this paper, a simple IIR filter is used in system identification. A uniform white sequence is used as an input signal for the unknown system. A noise white sequence signal, which is not correlated with the input signal, is added to the system output. An illustrative example is solved, and the optimization is performed using the Simplex method by Nelder and Mead. A comparison is done to results by a genetic algorithm and a simulated annealing algorithm. It is demonstrated, that a gradient-based algorithm gets stuck in a local minimum. The obtained result confirms the efficiency and efficacy of this approach. 

Adaptive IIR Filtering for System Identification applying the method by Nelder and Mead

"Atrial fibrillation (AF) is a cardiovascular disease identified as one of the main risks
factors for stroke. The majority of strokes due to AF are caused by clots originating in the left atrial appendage (LAA). LAA occlusion is an effective procedure for reducing stroke risk.  Planning the procedure using pre-procedural imaging and analysis has shown benefits. The analysis is commonly done by manually segmenting the appendage on 2D slices. Automatic LAA segmentation methods could save an expert's time and provide insightful 3D visualizations and accurate automatic measurements to aid in medical procedures. Several semi- and fully automatic methods for segmenting the appendage have been proposed. This paper provides a review of automatic LAA segmentation methods on 3D and 4D medical images, including CT, MRI, and echocardiogram images. We classify methods into heuristic and model-based methods, as well as semi- and fully-automatic methods. We summarize and compare the proposed methods, evaluate their effectiveness, present current challenges in the field and approaches to overcome them."

Premium content

Detection of optic disc and macula in retinal images using deep learning methods

Downloads

Next from IWSSIP 2021

Gait Recognition with DensePose Energy Images

Similar lecture

SFPN: Semantic Feature Pyramid Network for Object Detection