Search

We introduce AXS (Astronomy eXtensions for Spark), a scalable open-source astronomical data analysis framework built on Apache Spark, a state-of-the-art industry-standard engine for big data processing. In the age when the most challenging questions of the day demand repeated, complex processing of large information-rich tabular datasets, scalable and stable tools that are easy to use by domain practitioners are crucial. Building on capabilities present in Spark, AXS enables querying and analyzing almost arbitrarily large astronomical catalogs using familiar Python/AstroPy concepts, DataFrame APIs, and SQL statements. AXS supports complex analysis workflows with astronomy-specific operations such as spatial selection or on-line cross-matching. Special attention has been given to usability, from conda packaging to enabling ready-to- use cloud deployments. AXS is regularly used within the University of Washington's DIRAC Institute, enabling the analysis of ZTF (Zwicky Transient Facility) and other datasets. As an example, AXS is able to cross-match Gaia DR2 (1.7 billion rows) and SDSS (710 million rows) in 25 seconds, with the data of interest (photometry) being passed to Python routines for further processing. Here, we will present current AXS capabilities, give an overview of future plans, and discuss some implications to analysis of LSST and similarly sized datasets. The long-term goal of AXS is to enable petascale catalog and stream analyses by individual researchers and groups.

In this work, the U-Net neural network with a custom postprocessing method has been applied for automatic segmentation of retinal layers from a database of OCT images with age-related macular degeneration (AMD).

Purpose: Fully automated and accurate quantification of retinal images is emerging as a necessary tool for the clinical utility of optical coherence tomography (OCT) images. The purpose of this study is to investigate machine learning, in particular Deep Neural Networks (DNN), to perform the classification and segmentation necessary for quantification of microvasculature in OCT Angiography images, fluid in OCT Bscans, and photoreceptors in Adaptive Optics (AO) OCT images. Methods: Each of the automated processing networks required different methods. For microvasculature segmentation, OCTA images of both healthy and diabetic eyes were acquired with a prototype and two commercial systems. The DNN was trained on expert segmentations, and the output was used to calculate vessel density as well as four foveal avascular zone morphometric parameters (area, maximum and minimum diameter, and eccentricity). For the segmentation and classification of retinal fluid, the images used were acquired from commercially available systems. The processing included layer segmentation, fluid detection, and classification as intraretinal fluid (IRF), subretinal fluid (SRF) and pigment epithelial detachment (PED). Lastly, the images used for quantification of photoreceptors were acquired with a recently developed Sensorless AO-OCT with a 200kHz swept source OCT engine. Results: For the microvasculature segmentation, the average pixel- wise accuracy was 0.823 across all systems. Additionally, no significant difference between the means of the measurements from automated and manual segmentations were found for any of the clinical outcome measures on any system. The proposed fluid segmentation network achieved a Dice index of 0.767 for segmentation and an Area Under the Curve (AUC) measure of 1.00 for detection tasks. For the cone counting algorithm, the accuracy of the DNN segmented output was found to be 0.981. Representative OCT images before and after segmentation Figure 1 (a) – (f). Figure 1 (d) is the probability map output of the DNN. Figure 1 (e) shows the fluid segmentations and classification as IRF (red), SRF (yellow) and PED (blue) along with the anterior and posterior segmentations of the retina. Figure 1 (f) shows the result of the DNN where each cyan mark is noted as a center of a cone. Conclusions: Machine learning based segmentation enables reliable quantification of retinal OCT scans.

A method for traffic sign shape detection is presented, which does not require extensive training and is based on fast template matching suitable for real-time applications. The proposed method is able to detect prohibitory and warning signs. Post processing phase enables detection verification thus drastically increasing detection accuracy. All method parameters, if required, can be tweaked without the need for time-consuming retraining procedures. The method shows robustness to various conditions that often impose problems for color based systems. The proposed method can be used standalone or with already developed methods as a verification layer.

Our approach for detecting fiber orientation in volumes acquired by slice-wise imaging is to use well known methods for tracking objects in videos, based on the brightness constancy principle: as in the Horn-Schunck and Lucas- Kanade algorithms. We merge the vector fields obtained along each of the three directions with these two algorithms into one resulting vector field to compensate for the limitations of each algorithm. Though brightness constancy approaches are commonly used in video applications, we managed to obtain good results on static 3D images by combining the original approaches.

Knowledge about the microstructure is crucial in targeted synthesis of novel nanomaterials. The microstructural parameters, crystallite size and crystallite strain play a major role in physical and chemical properties of the material. X-ray diffraction (XRD) is a very suitable method for this task, since it is non-destructive and it enables a very quick and precise determination of these parameters. The main problem lies in the case where the two neighboring diffraction profiles overlap each other. Here we present a new method for the separation of the overlapping profiles based on the differentiation of the profiles. Further, this method is appropriate for non- crystallographers working in the field of material science since it does not require any crystallographic experience and the full knowledge about the structure of the sample investigated. The microstructural results obtained by the proposed method are very accurate.

Complex helical organization of cardiac fibers is one of the key factors for efficient beat- to- beat contraction and electrical impulse propagation. Complete understanding of this (inter-individual) configuration is limited by image acquisition and analysis constraints. Consequently, extensive quantification of myofiber orientation and remodeling within diverse cases is still lacking. With its high resolution and contrast, synchrotron-based phase- contrast X-ray imaging offers potential for assessing this information. Although it recently gained increased attention for biomedical purposes, only few cardiac applications were presented to this date. In this paper, we used synchrotron-based acquisitions of a healthy fetal rabbit heart and implemented a structure tensor method for estimating fiber orientation. For comparison, we generated the common rule-based model, simulating fiber angles distribution for the given geometry. Although we find similar fiber angle transmural courses compared to the theoretical, high-resolution imaging and analysis show that the myocardium in an individual is more complex than often assumed.

Traffic accidents are more frequent during night- time despite lower traffic volume. Driver assistance systems can increase car and road safety. Such systems are often visual based, where the base is vehicle light detection. A method for night-time vehicle light detection is presented which is capable to detect front and rear vehicle lights. The detection procedure operates on raw pixels, which does not need extensive image pre- processing such as image segmentation thus improving robustness and performance. Although two classes of vehicle lights are detected, a single binary classifier architecture is used to detect all categories of vehicles lights. The obtained detections are then labeled by using the same approach. The result is a simple yet effective architecture capable to detect vehicle lights in various night-time lighting conditions.

Images of fetal rabbit hearts affected with intrauterine growth restriction IUGR (acquired in ESRF), are computationally examined, with a focus on cardiac fiber orientation and coronary vasculature detection. With different fiber orientation and vasculature comparing to the healthy hearts, we imply that the specific cardiac remodeling occurs as a result of the heart's adaptation to IUGR.

A novel method for vehicle detection and tracking, which is capable to correctly detect and track vehicle headlights in rural and urban areas is presented. The detection procedure does not require image thresholding, or other preprocessing, which is commonly used in other methods, thus offering robustness and performance. A tracking method based on Joint Probability Data Association Filter (JPDAF) efficiently associates object detections with existing tracks and provides a track management system. The method deals with object appearance and disappearance and can handle false detections introduced by the frame-based detector.

In this paper an improvement of the white patch method, a color constancy algorithm, is proposed. The improved method is tested on several benchmark databases and it is shown to outperform the baseline white patch method in terms of accuracy. On the benchmark database it also outperforms most of the other methods and its great execution speed makes it suitable for hardware implementation. The results are presented and discussed and the source code is available at http://www.fer.unizg.hr/ipg/resources/color_constancy/.

Vision is the most important sense used for driving and therefore computer vision algorithms are the most critical for Advanced Driver Assistance System (ADAS). In this paper, we present an overview of the current research results of the Image Processing Group in the area of traffic-related computer vision methods. First, we present a method for headlight and taillight detection, which can be used in systems for headlight control and forward collision warning. Second, we present a method for detection of roadside vegetation, which can be used for roadside maintenance and detection of vegetation- like obstacles in traffic. Computational color constancy is important for computer vision systems to achieve invariance to scene illumination. Third, a method for achieving color constancy is presented in this paper.

Systems for vehicle surround view are increasingly used for driver assistance systems in automotive industry. Such systems assists the driver by displaying a view of the vehicle surroundings on the LCD display placed on the dashboard. This allows for easier, quicker and more controllable maneuvers while parking. In this paper, we present a novel system for real- time surround view vehicle visualization using four fish-eye cameras mounted on the vehicle. We propose novel methods for geometric calibration of individual cameras and for calibration of the whole camera system, once the cameras are mounted on the vehicle. The proposed calibration methods have been designed with industrial production in mind. Experimental results have shown that the proposed calibration methods are able to perform an accurate calibration of individual cameras and the complete camera system. The described system is implemented using FPGA hardware and is able to perform complex image warping and stitching operations in real-time.

Large impact event about 65 million of years ago redirected biological evolution on Earth by producing the major extinction – the point at which dinosaurs died and mammals became major players in the history of terrestrial life, while the multiple impacts of Comet Shoemaker-Levy 9 on Jupiter in July 1994 provided an additional example of what might happen if a similar object struck Earth instead [B. M. French, 1998, Traces of Catastrophe]. According to the Earth Impact Database, there are 178 confirmed impact craters on Earth (EA178 catalogue). Estimates of yet-to-find impact crater population on Earth depend on catalogues of impact craters of other planetary bodies, such as Mars, Venus and the Moon, wherein considerably more craters are preserved up to the present date [S. A. Stewart, 2011, Journal of the Geological Society 168 (1), 1-13]. Scientific investigation of all these processes is important for prediction of possible future impacts on Earth and consequential natural hazards. In our previous work, we developed a crater detection algorithm (CDA) for crater detection from digital topography data using gradient value/orientation, morphometry, votes-analysis, slip-tuning and calibration [G. Salamunićcar and S. Lončarić, 2010, IEEE Transactions on Geoscience and Remote Sensing 48 (5), 2317-2329] and further improved it using a crater shape-based interpolation [G. Salamunićcar, S. Lončarić and E. Mazarico, 2012, Planetary and Space Science 60 (1), 236-247]. During the work on some of the currently most complete catalogues of Mars (MA132843GT with 132843 craters) and the Moon (LU60645GT with 60645 craters), we used CDA to identify numerous previously uncatalogued craters and extended the catalogues with them. In this work, we are using our CDA in combination with Earth digital elevation map (DEM) dataset ACE2 (Altimeter Corrected Elevations) in horizontal resolution of 30 minutes (1/120°) upsampled into 1/128°, in order to evaluate the possibility of potential search for still unknown impact structures on Earth. Out of 178 known impact craters (EA178), our CDA successfully detected 24 of them (other non detected craters are mostly too eroded or too small for available resolution). This confirms that our CDA can be used successfully for detection of Earth impact craters as well. The plan for our future work is to use DEM datasets in much higher horizontal resolution (e.g. ACE2 available in horizontal resolution of 3 minutes – 1/1200°), in order to search for still unknown impact craters on Earth.

Detecting changes in the contractility of the heart muscle, especially in the presence of coronary artery disease, is an important medical task. To test whether there is a relationship between the morphology of the aortic outflow velocity profile and myocardial function a segmentation of the aortic outflow velocity profile is necessary as well as the morphology feature extraction. Our approach is based on atlas-based image segmentation that utilizes specially formulated geometrical transformation, image similarity measure, atlas formation and segmentation propagation. Following atlas-based image segmentation, various cardiac parameters are extracted to describe morphological properties of transaortic valvular flow.

We used a crater detection algorithm for detection of craters from Venus digital topography and computation of the depth/diameter ratio. The results were compared with the accompanying results for Martian and Lunar craters.