|
|
Deep learning-based noise reduction in X-ray images
Říhová, Barbora ; Jakubíček, Roman (oponent) ; Zemek, Marek (vedoucí práce)
X-ray imaging technology is the foundation for exploring the internal structure of a wide range of objects, however the results can be compromised by noise. This thesis is focused on the removal of noise in X-ray projections using deep learning, that has the capability to adapt to a specific task. The thesis contains a theoretical investigation focusing on the areas of X-ray production and detection, noise in X-ray images, and neural networks. A special chapter is devoted to the description of the chosen solution, which is performed by creating a dataset partially consisting of modeled X-ray projections with the subsequent incorporation of noise corresponding to noise model in real images and partly from X-ray projection series. The RIDNet convolutional neural network architecture was selected for implementation, since it shows good result for denoising task. Three models were trained using different parts of the dataset. The best performance was observed for models, that used real data for training. Their performance is comparable to traditional methods such as BM3D.
|
|
|
Automating the search for abnormalities from tomographic data
Semerák, Petr ; Zemek, Marek (oponent) ; Bazala, Jiří (vedoucí práce)
The presented thesis concerns the automation of the process of scanning and evaluating abnormalities of gascoolers that occur during their production. Non-destructive testing employs CT scanning technology, which generates image data as an output. The objective of the work is to replace the time-consuming manual data scanning process with a reliable algorithmic method and to assess the potential of this direction of development. The theoretical part of the thesis deals with cooling systems in cars, non-destructive testing with a focus on CT technology and a search for software for viewing and analysing CT data. The practical part of the thesis focuses on the problem of clogged gascooler ducts. The causes of this abnormality, the current approach to its detection and a new automatic inspection approach are described. The proposed algorithm together with an application developed using Matlab are tested on concrete data. Finally, the reliability of the results is evaluated by manual inspection of the CT images. A deep neural network is trained to assess the quality of the image data.
|
|
|
Method for Extending the Field of View for X-ray Computed Tomography with Submicron Resolution
Zemek, Marek ; Chmelík, Jiří (oponent) ; Mézl, Martin (vedoucí práce)
Computed tomography is a tool for nondestructive evaluation of samples, commonly used in many industrial and scientific fields. Some tomographic devices produce images with sub-micrometer spatial resolution. The field of view of such devices can be very small, in the range of single millimeters or less. This restricts possible sizes of samples, which is a major limitation. Various field-of-view extension techniques exist which are able to overcome this restriction. In this thesis, a previously published technique was adapted and implemented specifically for the Rigaku Nano3DX X-Ray microscope. This technique almost doubles the lateral extent of the field of view without the need for a larger detector array. The approach was tested using both synthetic and real data, and its performance is evaluated subjectively and objectively, through visual inspection and image quality metrics. The evaluation is largely based on comparing images reconstructed using this method to ones acquired using a larger detector array. The field-of-view extension method yields faithful reconstructions of samples, comparable in quality to their larger-detector counterparts.
|
|
|
Algoritmy pro rekonstrukci obrazu z projekcí
Zemek, Marek ; Chmelík, Jiří (oponent) ; Mézl, Martin (vedoucí práce)
Práce je zaměřena na problematiku rekonstrukce obrazu z jeho projekcí a na hodnocení kvality této rekonstukce pomocí různých typů algoritmů. Obzvláštní pozornost je věnována aplikaci těchto algoritmů v lékařské zobrazovací modalitě CT. Úvodní část se zabývá teoretickým rozborem algoritmů používaných v současnosti. Následně je popsána realizace několika jednoduchých rekonstrukčních metod v prostředí Matlab, jejich aplikace na simulovaná i reálná data a následné zhodnocení a porovnání z hlediska kvality rekonstrukce a výpočetní náročnosti.
|
|
|
Method For Extending The Field Of View For X-Ray Computed Tomography With Submicron Resolution
Zemek, Marek
Computed tomography allows for nondestructive evaluation of samples. It is commonly used for many industrial and scientific applications. Some devices are capable of submicron resolutions, but this often comes at the cost of a limited field of view. Techniques that extend the field of view can greatly enhance the versatility of these scanners. One such technique is presented here. It is implemented on the Rigaku Nano3DX, almost doubling its lateral field of view. The method utilizes a standard reconstruction algorithm, and yields faithful reconstructions of scanned samples without the need for a larger detector.
|
|
|
Method for Extending the Field of View for X-ray Computed Tomography with Submicron Resolution
Zemek, Marek ; Chmelík, Jiří (oponent) ; Mézl, Martin (vedoucí práce)
Computed tomography is a tool for nondestructive evaluation of samples, commonly used in many industrial and scientific fields. Some tomographic devices produce images with sub-micrometer spatial resolution. The field of view of such devices can be very small, in the range of single millimeters or less. This restricts possible sizes of samples, which is a major limitation. Various field-of-view extension techniques exist which are able to overcome this restriction. In this thesis, a previously published technique was adapted and implemented specifically for the Rigaku Nano3DX X-Ray microscope. This technique almost doubles the lateral extent of the field of view without the need for a larger detector array. The approach was tested using both synthetic and real data, and its performance is evaluated subjectively and objectively, through visual inspection and image quality metrics. The evaluation is largely based on comparing images reconstructed using this method to ones acquired using a larger detector array. The field-of-view extension method yields faithful reconstructions of samples, comparable in quality to their larger-detector counterparts.
|
|
|
Algoritmy pro rekonstrukci obrazu z projekcí
Zemek, Marek ; Chmelík, Jiří (oponent) ; Mézl, Martin (vedoucí práce)
Práce je zaměřena na problematiku rekonstrukce obrazu z jeho projekcí a na hodnocení kvality této rekonstukce pomocí různých typů algoritmů. Obzvláštní pozornost je věnována aplikaci těchto algoritmů v lékařské zobrazovací modalitě CT. Úvodní část se zabývá teoretickým rozborem algoritmů používaných v současnosti. Následně je popsána realizace několika jednoduchých rekonstrukčních metod v prostředí Matlab, jejich aplikace na simulovaná i reálná data a následné zhodnocení a porovnání z hlediska kvality rekonstrukce a výpočetní náročnosti.
|
host ::
RSS.