Kenar bulma algoritması ve regresyon ile dct-tabanlı referanssız bloklanma ölçüm metodu

This thesis presents a blocking artifact detection algorithm and consists of detailed experimental results. Blocking artifacts are a serious problem in low data rate transmission of such videos (or image sequences), which employ DCT-based compression algorithms. Too much compression due to bandwidth constraints can introduce blocking artifacts. Online quality measurement techniques for artifacts such as blockiness are required to for broadcasters and internet service providers to inspect their streaming systems. Blockiness measurement can be done through full-reference (FR), reduced-reference (RR), or no-reference (NR) methods. NR is the most efficient one as accessing reference content may not be always possible. NR methods also offer the possibility of real-time inspection. We propose a novel NR blockiness measurement method. The method is based on DCT and uses a model of the human visual system like some of the literature. However, it is unique as it uses Sobel edge detection algorithm to avoid miscalculations and also employs regression analysis to match mean opinion scores (MOS) of human observers. These unique features of our method let us achieve results that are more correlated to the MOS for LIVE dataset, which is used to train and test our algorithm. To measure how well our results match the MOS values in LIVE, we use Pearson and Spearman correlation formulations. The best Spearman correlation obtained in the literature for LIVE dataset is 94% while we get 95%. On the other hand, we get a Pearson correlation of 98% and surpass the best in the literature by 94%.