New Fast Block Matching Algorithm Using New Hybrid Search Pattern And Strategy To Improve Motion Estimation Process In Video Coding Technique

• 第一著者: Hamid, Nurul 'Atiqah
• フォーマット: 学位論文
• 言語: 英語; 英語
• 出版事項: 2016
• 主題: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• オンライン･アクセス: http://eprints.utem.edu.my/id/eprint/20604/1/New%20Fast%20Block%20Matching%20Algorithm%20Using%20New%20Hybrid%20Search%20Pattern%20And%20Strategy%20To%20Improve%20Motion%20Estimation%20Process%20In%20Video%20Coding%20Technique.pdf; http://eprints.utem.edu.my/id/eprint/20604/2/New%20Fast%20Block%20Matching%20Algorithm%20Using%20New%20Hybrid%20Search%20Pattern%20And%20Strategy%20To%20Improve%20Motion%20Estimation%20Process%20In%20Video%20Coding%20Technique.pdf

Up until today, video compression algorithm has been applied in various video applications ranging from video conferencing to video telephony. Motion Estimation or ME is deemed as one of the effective and popular techniques in video compression. As one of its techniques, the Block Matching Algorithm or BMA is widely employed in majority of well-known video codes due to its simplicity and high compression efficiency. As such, it is crucial to find different approaches of fast BMAs as the simplest and straightforward BMA is not a good fit for implementation of real-time video coding because of its high computational complexity. The aims for this study is to develop and design a new hybrid search pattern and strategy for new fast BMAs that can further improve the ME process in terms of estimation accuracy and video image quality, searching speed and computational complexity. There are 6 main designs that the algorithms proposed namely the Orthogonal-Diamond Search Algorithm with Small Diamond Search Pattern (ODS-SDSP), the Orthogonal-Diamond Search Algorithm with Large Diamond Search Pattern (ODS-LDSP), the Diamond-Orthogonal Search Algorithm with Small Diamond Pattern (DOS-SDSP), the Diamond-Orthogonal Search Algorithm with Large Diamond Pattern (DOS-LDSP), the Modified Diamond-Orthogonal Search Algorithm with Small Diamond Pattern (MDOS-SDSP), and the Modified Diamond-Orthogonal Search Algorithm with Large Diamond Pattern (MDOS-LDSP). These 6 algorithms are divided into 3 main methods namely Method A, Method B, and Method C depending on their search patterns and strategies. The first method involves the manipulation of the diamond pattern in the process, the second method includes the manipulation of the orthogonal steps, and lastly, the third method is the modified version of the second method to improve the performances of the algorithms. Evaluation is based on the algorithm performances in terms of the search points needed to find the final motion vector, the Peak-Signal to Noise Ratio (PSNR) of the algorithms, and the runtime performance of algorithm simulations. The result shows that the DOS-SDSP algorithm has the lowest search points with only 1.7341, 4.9059 and 4.0230 for each motion’s content respectively; meanwhile all the algorithms acquired similar and close PSNR values for all types of motion contents. As for simulation runtime, the results show that Method B has the least simulation runtime and Method C has the highest simulation runtime compared to others for all video sequences. The finding suggests that an early termination technique should be implemented at the early stage of the process, and mixing the selection of the mode is able to improve the algorithm performances. Therefore, it can be concluded that Method B gives the best performance in terms of search points reduction and simulation runtime while Method C yields the best for PSNR values for all types of motion contents.