|
|
GPU Raytracer
Herrmann, Pavel ; Pelikán, Josef (advisor) ; Horáček, Jan (referee)
Ray tracing is a popular method for generating realistic imagery, with high computation complexity and high potential for parallelization. Modern GPUs can be used as a high performance parallel co-processor, making them seemingly ideal for tasks such as ray tracing. This thesis will give an overview of ray tracing methods, overview of GPU computing methods, and present a piece of software designed for using GPU for ray tracing. This software tries to integrate classic ray tracing algorithms with specifics of GPU programming, while providing extendability and high performance.
|
|
|
Mesh mending
Masařík, Tomáš ; Pelikán, Josef (advisor) ; Krajíček, Václav (referee)
The bachelor work is focused on triangle meshes mending. It includes topological and also geometrical aspects. I continue working on Individual Software Project and Software Praxis. Before I have created program capable of detecting and repairing some of triangle mesh errors. Main goal of creation such a program was a research using scans of human faces at Faculty of Anthropology of Nature Science at Charles University in Prague. There are analyzed issues and there are evaluated existing solution in this paper. In second section I describe implemented algorithms, user's and programmer's documentation as well. The results are demonstrated At the end and there are proposed some future extensions.
|
|
|
Point Primitives on GPU
Skotáková, Jana ; Pelikán, Josef (advisor) ; Kmoch, Petr (referee)
In recent years representation of objects based on point primitives receives more interest. This thesis presents possibilities of processing of point-sampled models. It gives an overview of algorithms that can be used in solving problems like deformation or animation of point-sampled models. There is also described how such models can be obtained. In the following part Level Of Detail is presented with regard to point-sampled models. There are given basic methods and factors used for the LOD selection. Furthermore, there is proposed and implemented transfer of parts of the selected LOD generating algorithm to GPU. This keeps CPU available for other parts of the algorithm or even other tasks. DirectX 10 is used for the GPU programming. DirectX 10 uses the new model of graphic pipeline that enables the transfer of more parts of algorithm to GPU.
|
|
|
L-systems online
Fišer, Marek ; Pelikán, Josef (advisor) ; Mráz, František (referee)
An L-system in its simplest form is a variant of a context-free grammar. Originally, L-systems were developed and are still mainly used for modeling plant growth, though with L-systems it is possible to create general fractals, models of towns or even music. However, anyone interested in L-systems and wanting to experiment with them may have difficulty finding an appropriate application. The goal of this work was to create an online system, suitable for a wide range of users, for working and experimenting with L-systems. The resulting solution consists of two parts. The first part is a universal, easily-expandable library for processing L-sys- tems. Expandability is achieved thanks to its modularity. All input is processed through interconnected components that are specialized in particular activities. The specialization of the components also contributes to the clarity and reliability of the whole processing system. The library is independent and multiplatform and can thus be readily used in other applications. The second part consists of a modern web interface designed to be under- standable for beginners and yet also capable of providing advanced features for more advanced users. Part of the site is a gallery of L-systems to which each user can contribute and which thus helps to create a user-community. The...
|
|
|
Modelling of deformations of geometric objects
Kratochvíl, Erik ; Kolingerová, Ivana (advisor) ; Pelikán, Josef (referee)
Deformable models have been widely studied by the CG community for more than two decades. Many issues had to be addressed and many problems had to be solved before the quality of the deformable models reached an acceptable level. The aim of the work is to simulate interactions of several deformable bodies in realtime. First, we unveil the basic ideas behind several deformable models created for solids represented by a surface or volumetric mesh. We prefer the physicallybased approaches as they tend to yield more convincing results. We consider elastic materials only. We also briefly discuss the topic of collision detection for deformable models with its speci c aspects. A special attention is paid to contact resolution because it greatly influences the final impression. The result of this thesis is an overview of the proposed algorithm, detailed description of its parts, and its implementation. We also perform several benchmarks to prove its applicability in virtual environments and its capability to run in realtime.
|
|
|
Zpracování medicínských dat na GPU
Hlaváček, Jakub ; Maršálek, Lukáš (advisor) ; Pelikán, Josef (referee)
The aim of this thesis is to design and experimentally implement a complex framework dealing with accelerating and simplifying the development of systems for processing and visualization of medical volume data in C#. Currently, there are application interfaces and their implementations for both, techniques based on image processing, like filtering, registration, segmentation and classification, and also for techniques based on 3D image visualization. But there is no consistent framework for both tasks, which would take advantage of features of modern graphics processing units and multi-core processing units along with features of .NET Framework and of language C#. The thesis presents overview of current free and open source software, design of application interface, and implementation of main API features. One of important differences to other software is that the implementation has been developed natively in the managed environment of .NET Framework, offering a good level of comfort for an end application programmer, but system performance is comparable with natively compiled environments thanks to utilization of all resources.
|
|
|
Video Watermarking
Zlomek, Martin ; Pelikán, Josef (advisor) ; Doubek, Petr (referee)
Nazev prace: Watermarking videa Autor: Martin Zlomek Katedra (ustav): Kabinet software a vyuky informatiky Vedouci diplomove prace: RNDr. Josef Pelikan e-mail vedouci'ho: josef.pelikan@mff.cuni.cz Abstrakt: Navrzeny a implementovany byly tfi ruzne metody pro watermarking video sekvencf, ktere jsou komprimovany podle standardu H.264. Dve z techto metod reprezentujf metody, ktere watermark vkladaji ve frekvencni oblasti, zati'mco tfeti path' mezi metody, ktere watermark vkladaji do obrazove oblasti. Vkladani watermarku ve frekvencni oblasti probiha zmenou transformacnich koeficientu, ktere jsou ziskany pri'mo z komprimovaneho proudu dat. Watermark, ktery ma byt vlozen v obrazove oblasti, je pred vlozenim do techto koeficientu nejprve transformovan do frekvencnioblasti. Dale byl navrzen a implementovan obecny watermarkovaci system, ktery poskytuje jednoduche rozhrani usnadnujici implementaci konkretnich metod. Odolnost navrzenych metod vuci ruznym utokum byla provefena a vzajemnc porovnana sadou nekolika testu. Testy simuluji nasledujici utoky: rekompresi, zmenu mefitka, ofezani, zbaveni sumu, zasumeni, rozmazani, zaostfeni, mnohonasobne vkladani watermarku a tzv. konspiracni utok. S ohledem na robustnost a viditelnost watermarku v obraze je metoda, ktera watermark vklada do obrazoveoblasti, preferovana pfed...
|
|
|
Real-time visualization of terrain data
Šrámek, Ondřej ; Maršálek, Lukáš (advisor) ; Pelikán, Josef (referee)
The rendering of huge terrain data is an actual problem in many application areas like interactive games or virtual environments. To keep rendering interactive it was necessary to implement computationally intensive LOD techniques, because the graphical systems were incapable of rendering enough geometry to represent the whole terrain accurately. Due to the raw processing power of current systems (with respect to rasterization) which enables to render hundreds of millions of triangles per second, new methods for rendering these data emerge. They are based on observation that this is enough to render pixel-sized triangles while keeping reasonable output resolution. The aim of this work was to develop a practical implementation based on such approach. The main algorithm chosen was Geometry Clipmaps and its modi ed version was created. The solution devoloped is capable of rendering almost 300M samples terrain in the real-time with avarage 56 frames per seconds.
|
|
| |
|
|
Segmentation of neurocranium vessels for CT angiography
Tůma, Tomáš ; Pelikán, Josef (advisor) ; Maršálek, Lukáš (referee)
One of the key tasks involved in processing of CT angiography image data is the vascular system segmentation. The aim is to provide an objective vizualization of acquired data and thus make the diagnostics easier. This work aims at studying the methods used for segmentation of neurocranium vessels and implementing one of the existing algorithms in co-operation with radiodiagnostic department of Na Bulovce faculty hospital.
|
guest ::
RSS feed.