|
|
Methods and Tools for Image and Video Quality Assessment
Slanina, Martin ; Říčný, Václav (vedoucí práce)
The doctoral thesis is focused on methods and tools for image quality assessment in video sequences, which is a very up-to-date theme, undergoing a rapid evolution with respect to digital video signal processing, in particular. Although a variety of metrics for objective (automated) video sequence quality measurement has been developed recently, these methods are mostly based on comparison of the processed (damaged, e.g. with compression) and original video sequences. There are very few methods operating without reference, i.e. only on the processed video material. Moreover, such methods are usually analyzing signal values (typically luminance) in picture elements of the decoded signal, which is hardly applicable for modern compression algorithms such as the H.264/AVC as they use sophisticated techniques to remove compression artifacts. The thesis first gives a brief overview of the available metrics for objective quality measurements of compressed video sequences, emphasizing the different approach of full-reference and no-reference methods. Based on an analysis of possible ideas for measuring quality of video sequences compressed using modern compression algorithms, the thesis describes the design process of a new quality metric for video sequences compressed with the H.264/AVC algorithm. The new method is based on monitoring of several parameters, present in the transport stream of the compressed video and directly related to the encoding process. The impact of bitstream parameters on the video quality is considered first. Consequently, an algorithm is designed, employing an artificial neural network to estimate the peak signal-to-noise ratios (PSNR) of the compressed video sequences -- a full-reference metric is thus replaced by a no--reference metric. Several neural network configurations are verified, reaching from the simplest to three-layer feedforward networks. Two sets of video sequences are constructed to train the networks and analyze their performance and fidelity of estimated PSNRs. The sequences are compressed using the H.264/AVC algorithm with variable encoder configuration. The final part of the thesis deals with an analysis of behavior of the newly designed algorithm, provided the properties of the processed video are changed (resolution, cut) or encoder configuration is altered (format of group of pictures coded together). The analysis is done on video sequences with resolution up to full HD (1920 x 1080 pixels, progressive)
|
|
|
Nástroje pro měření kvality videosekvencí bez reference
Zach, Ondřej ; Dostál, Petr (oponent) ; Slanina, Martin (vedoucí práce)
Diplomová práce se zabývá objektivními metrikami pro hodnocení kvality videa bez reference. Jsou zde krátce popsány všeobecné základy hodnocení kvality videosekvencí, dále jsou uvedeny základní předpoklady objektivních metrik. Hlavním zaměřením práce jsou no-reference přístupy. Práce se snaží popsat základní metody pro hledání rušení ve videosekvencích. Je uveden rozdíl mezi metrikami pracujícími v prostorové a spektrální oblasti. Dále je popsán návrh aplikace v prostředí Matlab pro objektivní testy kvality videosekvencí. Byla navržena a implementována metrika pro odhad PSNR pro H.264 kódovaná videa na základě informací z bitové toku. Závěrem jsme sestavili databázi videosekvencí a provedli objektivní testy, které byly srovnány s výsledky testů subjektivních.
|
|
|
Methods and Tools for Image and Video Quality Assessment
Slanina, Martin ; Říčný, Václav (vedoucí práce)
The doctoral thesis is focused on methods and tools for image quality assessment in video sequences, which is a very up-to-date theme, undergoing a rapid evolution with respect to digital video signal processing, in particular. Although a variety of metrics for objective (automated) video sequence quality measurement has been developed recently, these methods are mostly based on comparison of the processed (damaged, e.g. with compression) and original video sequences. There are very few methods operating without reference, i.e. only on the processed video material. Moreover, such methods are usually analyzing signal values (typically luminance) in picture elements of the decoded signal, which is hardly applicable for modern compression algorithms such as the H.264/AVC as they use sophisticated techniques to remove compression artifacts. The thesis first gives a brief overview of the available metrics for objective quality measurements of compressed video sequences, emphasizing the different approach of full-reference and no-reference methods. Based on an analysis of possible ideas for measuring quality of video sequences compressed using modern compression algorithms, the thesis describes the design process of a new quality metric for video sequences compressed with the H.264/AVC algorithm. The new method is based on monitoring of several parameters, present in the transport stream of the compressed video and directly related to the encoding process. The impact of bitstream parameters on the video quality is considered first. Consequently, an algorithm is designed, employing an artificial neural network to estimate the peak signal-to-noise ratios (PSNR) of the compressed video sequences -- a full-reference metric is thus replaced by a no--reference metric. Several neural network configurations are verified, reaching from the simplest to three-layer feedforward networks. Two sets of video sequences are constructed to train the networks and analyze their performance and fidelity of estimated PSNRs. The sequences are compressed using the H.264/AVC algorithm with variable encoder configuration. The final part of the thesis deals with an analysis of behavior of the newly designed algorithm, provided the properties of the processed video are changed (resolution, cut) or encoder configuration is altered (format of group of pictures coded together). The analysis is done on video sequences with resolution up to full HD (1920 x 1080 pixels, progressive)
|
|
|
Nástroje pro měření kvality videosekvencí bez reference
Zach, Ondřej ; Dostál, Petr (oponent) ; Slanina, Martin (vedoucí práce)
Diplomová práce se zabývá objektivními metrikami pro hodnocení kvality videa bez reference. Jsou zde krátce popsány všeobecné základy hodnocení kvality videosekvencí, dále jsou uvedeny základní předpoklady objektivních metrik. Hlavním zaměřením práce jsou no-reference přístupy. Práce se snaží popsat základní metody pro hledání rušení ve videosekvencích. Je uveden rozdíl mezi metrikami pracujícími v prostorové a spektrální oblasti. Dále je popsán návrh aplikace v prostředí Matlab pro objektivní testy kvality videosekvencí. Byla navržena a implementována metrika pro odhad PSNR pro H.264 kódovaná videa na základě informací z bitové toku. Závěrem jsme sestavili databázi videosekvencí a provedli objektivní testy, které byly srovnány s výsledky testů subjektivních.
|
host ::
RSS.