Slovakia

The COVID-19 (Coronavirus Disease 2019), caused by the SARS-CoV-2 virus, can cause severe pneumonia. The chest X-ray (CXR) is the preliminary radiographic exam to investigate pneumonia and other lung abnormalities. However, the identification of opacities in the CXR is challenging due to its overall low quality. In the literature, several works have proposed automated approaches to improve the quality of CXR images in general. However, as far as we know, few studies explored them in the context of COVID-19 identification in a thorough manner. In this work, we applied a series of digital image processing techniques, focusing on contrast enhancement. The impact of each method was evaluated using traditional digital image quality metrics, specifically peak signal-to-noise ratio (PSNR) and universal image quality (UQI), and a deep learning model for COVID-19 detection. We also investigated which regions were most frequently used by the model after each transformation using an Explainable Artificial Intelligence (XAI) approach called LIME. The rationale is to evaluate if any transformation improves the lung region used by the recognition model. The enhancement strategies that presented the overall best results considering PSNR and UQI were Gamma Correction (GC) and Adaptive Gamma Correction With Weighting Distribution (AGCWD). However, no method improved the identification of COVID-19 in a significant matter.

IWSSIP 2021

Chest X-ray image enhancement: a COVID-19 identification case study

chest x-ray

image enhancement

explainable artificial intelligence

covid-19


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)

technical paper

Clustering and classification techniques are frequently used to extract patterns and classify new data. Those techniques can be applied in data sets where the label information is not provided: the clustering interpretation identifies a pattern to assist the classification process. This interpretation often requires a specialist analysis since it depends on the approached problem but can also be performed by an automatic cluster labeling model. These models select the clusters' main characteristics to create labels that can be used as a summary of the clusters. In the literature papers in which these approaches are proposed, the labeling models present labels as attributes and ranges of values representing most clusters' elements. This paper proposes a cluster prediction model that uses cluster labels to generate rules and membership functions of a Fuzzy System to determine the cluster for new samples. The proposed model was evaluated by comparing accuracy, Kappa coefficient, and training time with other well-known classification algorithms. Also, we evaluated the number of rules generated. For different data sets used in the experiments, all available on the UCI repository, the results showed a Kappa coefficient above 0.8, which is considered excellent, and accuracy rate and training time similar to the well-known algorithms.

Cluster Prediction Model based on Automatic Cluster Labeling and Fuzzy System

An Automated Fingerprint Identification System (AFIS) is the cornerstone of many modern identity-driven applications, ranging from device authentication and law enforcement to security and borderline control. As the population in urban centers grows and the digitalization of services grows, so does the demand for more effective and efficient fingerprint recognition systems. Singular Points (SP), such as core and delta, are important landmarks that help to tackle this challenge. This paper proposes and evaluates an effective approach for SP detection based on a single-stage deep convolutional neural network model for object detection. We show that YOLOv4 detector with customized output layers is effective for handling cores and deltas patterns as patches in fingerprint images, using their center as coordinates. Experimental results were carried out on the challenging SPD2010 dataset to evaluate the proposed SP detector under different configurations. The best result is 60.34% of correctly detected fingerprints. In particular, compared to the state-of-the-art methods, our approach achieves an improvement up to 12% in correct detections, 8% in core detection rate, and 10% in delta detection rate. Core and delta miss rates are also reduced by 8% and 10%, respectively.

Robust Fingerprint Singular Point Detection using a Single-Stage CNN for Object Detection

 Cluster Labeling Models apply Artificial Intelligence techniques to extract the main characteristics of data partitioned into clusters to provide a tool for interpretation of the clustering. For this purpose, different techniques, such as Machine Learning, Fuzzy Logic, and Data Discretization are used to identify essential attributes for forming clusters and the ranges of values associated with them. This paper presents a Cluster Labeling Model based on regression that delimits the attributes ranges of values that best summarize the groups. In the proposed model, the importance of an attribute to clustering is determined by the capacity of its most accurate range of values -- according to regression-- to represent the clusters. The results achieved in the experiments have shown that the model is useful, providing specific labels for each cluster that represents between 84% and 100% of the elements of the clusters for the databases used.

Automatic Cluster Labeling Based on Regression Error Analysis

"Color is one of the main features for the evaluation of fruit ripeness.
Part of the quality control is based on the ripening status, where fruits should be
placed with others with the same ripening phase. However, this color analysis
can be done automatically using computer vision techniques. This work
presents an evaluation of the ripeness status of bananas, based on the analysis of
the texture features obtained with a digital camera. The results aim to determine
which texture feature can provide better results for the automatic classification
of the bananas ripening phase."

Banana Ripening Classification using Computer Vision: Preliminary Results

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.

Content-Based Multimodal Retrieval: Using Image and Audio Descriptors for Retrieving Cases from a Database of Cardiac Examinations

Static and Dynamic Point Cloud Compression by TSPLVQ

More than three quarters of world network data traffic is already video-related. This content does not only consist of professional and user generated clips, video from streamed and downloaded services and media pre-recorded or generated on the fly, but also content with different degrees of interactivity. Rich media also includes multi-sensorial media, omnidirectional video and mixed reality content. This content can be consumed at home, at work, in public places by a heterogeneous user population, while static or on the move. A wide range of devices contributes to generation and use of this content is being distributed over a network environment which is also highly heterogeneous. Challenges of transmitting this rich media content are highlighted and several solutions to support high quality, performance and/or energy efficiency of rich media delivery in such a complex context are presented.

Delivering High Quality Rich Media Content in Current Network Environment: Challenges and Solutions

JPEG AI: Towards the First Learning based Image Compression Standard

Detection and Localization of Objects within Images Using Computer Vision and Machine Learning

The paper focuses on three basic topics: an online environment for sharing of in-novative teaching materials; design and implementation of modern teaching mate-rials; and description of advanced types of learning objects used in such materi-als.
In these days we are about to conclude a set of projects that deal with develop-ment of highly innovative teaching materials and the environments needed for their delivery to the target groups. The most visible output of these projects is an online portal that provides access to hundreds of completely new teaching ma-terials containing a large number of learning objects.
In the paper we explain the technical specifics of the online environment, teaching materials and individual learning objects, which have a significant added value compared to conventional teaching aids.

Premium content

Chest X-ray image enhancement: a COVID-19 identification case study

Downloads

Next from IWSSIP 2021

Cluster Prediction Model based on Automatic Cluster Labeling and Fuzzy System

Similar lecture

New AIoT-Based Approach for Automatic Detection of COVID-19 Using Deep Learning Techniques in CT Images