|
|
Biometric fingerprint identification
Dašek, Filip ; Vítek, Martin (referee) ; Smital, Lukáš (advisor)
In biometrics we use distinctive physical features for identification and verification of identity. The most famous technique is identification by fingerprints. This technique use unique structure created by papillary lines for unambiguous identification. Thesis contains methods which were created throughout the years for analysis and adjustments of fingerprint. The algorithm is based on compairng two pairs of minitua and calculating transform matrix for correct alignment. Algorithm is tested on dataset created from LivDet databases. Performance of algorithm is represented by value EER which is compared with EERs of other algorithms tested in FVC 2006.
|
|
|
Fingerprints Generator
Chaloupka, Radek ; Orság, Filip (referee) ; Drahanský, Martin (advisor)
Algorithms for fingerprints recognition are already known for long time and there is also an effort for their best optimization. This master's thesis is dealing with an opposite approach, where the fingerprints are not being recognized, but are generated on the minutiae position basis. Such algorithm is then free of the minutiae detection from image and enhancements of fingerprints. Results of this work are the synthetic images generated according to few given parameters, especially minutiae.
|
|
|
Biometric fingerprint identification
Dašek, Filip ; Vítek, Martin (referee) ; Smital, Lukáš (advisor)
In biometrics we use distinctive physical features for identification and verification of identity. The most famous technique is identification by fingerprints. This technique use unique structure created by papillary lines for unambiguous identification. Thesis contains methods which were created throughout the years for analysis and adjustments of fingerprint. The algorithm is based on compairng two pairs of minitua and calculating transform matrix for correct alignment
|
|
|
Neural Network Implementation without Multiplication
Slouka, Lukáš ; Baskar, Murali Karthick (referee) ; Szőke, Igor (advisor)
The subject of this thesis is neural network acceleration with the goal of reducing the number of floating point multiplications. The theoretical part of the thesis surveys current trends and methods used in the field of neural network acceleration. However, the focus is on the binarization techniques which allow replacing multiplications with logical operators. The theoretical base is put into practice in two ways. First is the GPU implementation of crucial binary operators in the Tensorflow framework with a performance benchmark. Second is an application of these operators in simple image classifier. Results are certainly encouraging. Implemented operators achieve speed-up by a factor of 2.5 when compared to highly optimized cuBLAS operators. The last chapter compares accuracies achieved by binarized models and their full-precision counterparts on various architectures.
|
|
|
Fingerprint biometry
Smékal, Ondřej ; Drahanský, Martin (referee) ; Fedra, Petr (advisor)
Algorithms designed for identification and verification persons by fingerprints recognition are spread and used as in forensics aplications as in private sector for a long time. The aim of this thesis is to make us aquainted with various aplicated mathematic models of fingerprint processing in digital way. Second task is the presentation algorithmic solution of chosen subject identification procedure by force of Fingerprint matching. Algorithm is solid in the development environment platform Matlab.
|
|
|
Handwriting recognition using neural network
Petr, Martin ; Surynek, Pavel (advisor) ; Pergel, Martin (referee)
Title: Handwriting recognition using neural network Author: Martin Petr Department: Department of Theoretical Computer Science and Mathematical Logic Supervisor: RNDr. Pavel Surynek, PhD. Supervisor's e-mail address: pavel.surynek@mff.cuni.cz Abstract: Pattern recognition finds its use in many fields whose development has been affected by computer science and computer technology. Among these, the conversion of handwritten or printed text into computer-encoded text has a particularly prominent position. In the presented work we propose a method for recognizing handwritten characters in real-time using feedforward neural network as the basic classification mechanism. Dealing with differences among individual instances of each handwritten character we thoroughly explored the possibility of suppressing these while emphasizing characteristics that are essential for successful recognition. For these purposes we employed discrete cosine transform, whose time-proven application in audio and video signal processing or even directly in the field of pattern recognition provided a convincing argument for us to use it in our work as well. As a means of suppressing variations among various writing instruments we proposed preprocessing of input images using binarization and skeletonization. The designed method was...
|
|
|
Biometric fingerprint identification
Dašek, Filip ; Vítek, Martin (referee) ; Smital, Lukáš (advisor)
In biometrics we use distinctive physical features for identification and verification of identity. The most famous technique is identification by fingerprints. This technique use unique structure created by papillary lines for unambiguous identification. Thesis contains methods which were created throughout the years for analysis and adjustments of fingerprint. The algorithm is based on compairng two pairs of minitua and calculating transform matrix for correct alignment. Algorithm is tested on dataset created from LivDet databases. Performance of algorithm is represented by value EER which is compared with EERs of other algorithms tested in FVC 2006.
|
|
|
Biometric fingerprint identification
Dašek, Filip ; Vítek, Martin (referee) ; Smital, Lukáš (advisor)
In biometrics we use distinctive physical features for identification and verification of identity. The most famous technique is identification by fingerprints. This technique use unique structure created by papillary lines for unambiguous identification. Thesis contains methods which were created throughout the years for analysis and adjustments of fingerprint. The algorithm is based on compairng two pairs of minitua and calculating transform matrix for correct alignment
|
|
|
Neural Network Implementation without Multiplication
Slouka, Lukáš ; Baskar, Murali Karthick (referee) ; Szőke, Igor (advisor)
The subject of this thesis is neural network acceleration with the goal of reducing the number of floating point multiplications. The theoretical part of the thesis surveys current trends and methods used in the field of neural network acceleration. However, the focus is on the binarization techniques which allow replacing multiplications with logical operators. The theoretical base is put into practice in two ways. First is the GPU implementation of crucial binary operators in the Tensorflow framework with a performance benchmark. Second is an application of these operators in simple image classifier. Results are certainly encouraging. Implemented operators achieve speed-up by a factor of 2.5 when compared to highly optimized cuBLAS operators. The last chapter compares accuracies achieved by binarized models and their full-precision counterparts on various architectures.
|
|
|
Handwriting recognition using neural network
Petr, Martin ; Surynek, Pavel (advisor) ; Pergel, Martin (referee)
Title: Handwriting recognition using neural network Author: Martin Petr Department: Department of Theoretical Computer Science and Mathematical Logic Supervisor: RNDr. Pavel Surynek, PhD. Supervisor's e-mail address: pavel.surynek@mff.cuni.cz Abstract: Pattern recognition finds its use in many fields whose development has been affected by computer science and computer technology. Among these, the conversion of handwritten or printed text into computer-encoded text has a particularly prominent position. In the presented work we propose a method for recognizing handwritten characters in real-time using feedforward neural network as the basic classification mechanism. Dealing with differences among individual instances of each handwritten character we thoroughly explored the possibility of suppressing these while emphasizing characteristics that are essential for successful recognition. For these purposes we employed discrete cosine transform, whose time-proven application in audio and video signal processing or even directly in the field of pattern recognition provided a convincing argument for us to use it in our work as well. As a means of suppressing variations among various writing instruments we proposed preprocessing of input images using binarization and skeletonization. The designed method was...
|
guest ::
