|
|
Connection of algorithms for removal of influence of skin diseases on the process for fingerprint recognition
Heidari, Mona ; Derawi, Mohammad (oponent) ; Gomez-Barrero, Marta (oponent) ; Drahanský, Martin (vedoucí práce)
This thesis focuses on data structures, image processing, and computer vision methods for detecting and recognizing diseases in fingerprint images. The number of developed biometric systems and even used biometric characteristics is increasing. It is widely accepted that an individual's fingerprint is unique and remains relatively unchanged throughout life. However, the structure of these ridges can be changed and damaged by skin diseases. As these systems depend heavily on the structure of an individual's fingertip ridge pattern that positively determines their identity, people suffering from skin diseases might be discriminated against as their ridge patterns may be impaired. Likely, fingerprint devices have not been designed to deal with damaged fingerprints; therefore, after scanning the fingerprint, they usually reject it. The influence of skin disease is an important but often neglected factor in biometric fingerprint systems. An individual might be prevented from using specific biometric systems when suffering from a skin disease that affects the fingertips. Collecting a database of fingerprints influenced by skin diseases is a challenging task. It is expensive and time-consuming, but it also requires the assistance of medical experts and the ability to find willing participants suffering from various skin conditions on fingertips. The raw diseased fingerprint database is first analyzed to provide a solid foundation for future research. Common signs among all fingerprint images affected by the disease are found for every particular disease, and a general description of each disease and its influences is defined. Then we automatically assign the label based on a combination of the known state of the fingerprint image. The proposed solution is integrated with different algorithms focused on image processing libraries and computer vision methods for object detection. The solution has been evaluated on damaged fingerprint datasets and highlights the state of the art implementations using proposed techniques. The state of the art technique for disease detection implementations uses texture analysis and feature detection by comparing the intensity values of pixels in a small neighborhood in an image. Due to the complexity of each disease pattern, the combination of texture analysis algorithms leads to better detection results. The combination of GLCM, LBP, orientation field, and mathematical morphology can detect damage (artifacts) in fingerprint images. Combining these features makes it possible to identify changes in the texture and shape of the fingerprint flow caused by diseases. These techniques capture different aspects of the texture and shape of the damage in fingerprint images and lead to identifying changes in the texture caused by diseases. In the stages of the detection process, mathematical morphology operations are applied to improve the structural details by removing small irregularities in the image and simplify the shape of objects, making it easier to identify and isolate them expanding the boundaries of objects in an image or filling gaps and connect broken parts of objects, leading to better object detection and recognition. At the end of the detection process, coherence is applied to show the quality evaluation of fingerprint image patches into three types healthy, damaged, and background. Overall, the proposed solution showcases the effectiveness of integrating multiple image processing and computer vision algorithms for disease detection in fingerprint images.\ The combination of these algorithms can accurately detect and localize disease patterns in damaged fingerprint datasets, thus providing a reliable solution for disease detection in forensic applications.
|
|
| |
|
|
Generation of Synthetic Retinal Images with High Resolution
Aubrecht, Tomáš ; Heidari, Mona (oponent) ; Drahanský, Martin (vedoucí práce)
Special equipment, a fundus camera, is needed to capture the retina, which is the most important part of the human eye. Therefore, the main objective of this work is to design and implement a system that would be able to generate retinal images. The proposed solution uses an image-to-image translation, where the system is provided with a black and white image at the input containing only bloodstream, on the basis of which a color image of the entire retina is generated. The system consists of two neural networks: a generator, which generates retinal images, and a discriminator, which classifies these images as real or synthetic. Training of this system was performed on 141 images from publicly available databases. A new database was created with more than 2,800 images of healthy retinas in a resolution of 1024x1024. This database could be used as a learning tool for ophthalmologists or for the development of various applications working with retinas.
|
|
|
Indoor Detection of People Based on Vital Sign Sensing
Plaček, Dan ; Heidari, Mona (oponent) ; Drahanský, Martin (vedoucí práce)
Nowadays, in the era of smart solutions and a huge emphasis on data privacy, it is very desirable to detect the human presence anonymously and to recognize if the sensed person is in good health condition. This bachelor thesis is dealing with the detection of people based on vital sign sensing, specifically on the heart and respiratory rate sensing, an algorithmic approach is proposed, and the solution is implemented. The data are sensed via UWB pulse radar or FMCW radar, then the signal is processed, and vital signs are extracted. The thesis presents the experimental measurements and their results.
|
|
|
Detekce živosti otisku prstu na bezdotykovém zařízení
Fořtová, Kateřina ; Kanich, Ondřej (oponent) ; Heidari, Mona (vedoucí práce)
Tato bakalářská práce je zaměřena na detekci živosti otisků prstů s využitím bezdotykového senzoru. Shrnuje teoretický úvod do biometrie, zpracování otisků prstů a některé ze současných přístupů pro detekci živosti. Představuje nový přístup, který využívá algoritmus lokálního binárního vzoru, Sobelův a Laplaceův operátor a vlnkovou transformaci. Následná klasifikace byla provedena s využitím umělých neuronových sítí, metody podpůrných vektorů SVM a rozhodovacích stromů. Experimenty byly provedeny s datasetem nasvíceným světly o různé vlnové délce. Bylo zjištěno, že otisky prstů nasvícené červeným světlem vykazují nejlepší přesnost 90.1% ze všech uvažovaných vlnových délek viditelného světla. Klasifikace s využitím vektoru na základě lokálního binárního vzoru dosahovala průměrné přesnosti 89.8%, přesnost s užitým vektorem na základě Sobelova a Laplaceova operátoru byla 91.5%. Pro vlnkovou transformaci byly využity různé Wavelet rodiny. Největší přesnosti dosahovaly vlnky z rodiny biortogonálních spline vlnek (85.1%) a z rodiny reverzních biortogonálních spline vlnek (86.6%).
|
|
| |
|
|
Improvement of Methods for Detection and Classification of Damages in Fingerprint Images
Foltyn, Lukáš ; Heidari, Mona (oponent) ; Kanich, Ondřej (vedoucí práce)
This study aims to improve existing methods for detecting and classifying damage in fingerprint images by leveraging previous works conducted by students at Brno University of Technology. The work is built upon three applications: line damage (scars, hairs, creases) generator, moisture generator, and application containing multiple different models for fingerprint damage detection and classification. The three best-performing models - Faster-RCNN ResNet50, Faster-RCNN ResNet101, and CenterNet ResNet101 - were selected for further improvement. The work describes the creation of a dataset using undamaged synthetic fingerprint images, with the aforementioned damages introduced artificially. Efforts to improve the prediction accuracy of the models were based on more accurate annotation of bounding boxes and adjusting the hyperparameters. While the work yielded some improvements, the results are not consistently successful across all models and damage types.
|
|
|
Reconstruction and Enhancement of Damaged Parts of Fingerprint Images
Špila, Andrej ; Rydlo, Štěpán (oponent) ; Heidari, Mona (vedoucí práce)
This thesis deals with the problem of fingerprint image reconstruction with focus on non- recoverable regions affected by various skin diseases. A generative adversarial network with learnable convolutional gabor filter layer was trained on preprocessed dataset of real fingerprint images. The work demonstrates that the trained model can reliably repair small corrupted regions of arbitrary shapes and in case of larger holes, the global quality score of reconstructed fingerprints evaluated by MINDTCT module from NIST biometric image software is increased compared to original fingerprint. A standardized format for fingerprint images that helped stabilize the results when training generative models is proposed.
|
|
|
Indoor Detection of People Based on Vital Sign Sensing
Plaček, Dan ; Heidari, Mona (oponent) ; Drahanský, Martin (vedoucí práce)
Nowadays, in the era of smart solutions and a huge emphasis on data privacy, it is very desirable to detect the human presence anonymously and to recognize if the sensed person is in good health condition. This bachelor thesis is dealing with the detection of people based on vital sign sensing, specifically on the heart and respiratory rate sensing, an algorithmic approach is proposed, and the solution is implemented. The data are sensed via UWB pulse radar or FMCW radar, then the signal is processed, and vital signs are extracted. The thesis presents the experimental measurements and their results.
|
|
|
Generation of Synthetic Retinal Images with High Resolution
Aubrecht, Tomáš ; Heidari, Mona (oponent) ; Drahanský, Martin (vedoucí práce)
Special equipment, a fundus camera, is needed to capture the retina, which is the most important part of the human eye. Therefore, the main objective of this work is to design and implement a system that would be able to generate retinal images. The proposed solution uses an image-to-image translation, where the system is provided with a black and white image at the input containing only bloodstream, on the basis of which a color image of the entire retina is generated. The system consists of two neural networks: a generator, which generates retinal images, and a discriminator, which classifies these images as real or synthetic. Training of this system was performed on 141 images from publicly available databases. A new database was created with more than 2,800 images of healthy retinas in a resolution of 1024x1024. This database could be used as a learning tool for ophthalmologists or for the development of various applications working with retinas.
|
host ::
RSS.