Study Of Paired Comparison In Subjective Video Quality Assessment

Video quality evaluation standard plays a decisive role in the field of image and video processing, and is the benchmark to measure the performance of video-related processing algorithms, as well as optimizing video processing algorithms. Since 2005, A.C.Bovik put forward the defects of the mean square error in measuring image quality, and then the measurement method based on structural similarity. Academic study on evaluation method for image and video quality likes mushrooms to flourish. In recent years, the video quality evaluation methods have become a hot spot in image and video processing. This paper is supported by the National Natural Science Foundation for related research.The design goal of video quality evaluation method is to approach the subjective visual quality, thus in the algorithm design and validation, it is a must to subject to subjective tests based on real data in the database. The so-called real-world data, is based on the mean value of a series of subjective test experiments which is predetermined, therefore the validity of the results of subjective tests is critical in the chain. In smaller test collection database, it is easy to distinguish the quality of the data, so absolute classification method is often used for subjective tests. But such an approach is no longer applicable in the current test case of a sharp increase in the amount of data.In order to distinguish small differences between the measured data, international standard-setting organization adopts pair comparison method. But a full comparison for the paired comparison method, its time complexity is a growing by with the increase of data to be tested, so as to can not be achieved. In order to reduce complexity, scholars have put forward a variety of efficient method of paired comparison. Based on the analysis and study of the existing algorithms, starting with the time complexity problem, this paper improves a Ortho-exchange algorithm in reverse chronological order, and then proposes a paired comparison algorithm with more accurate results based on red-black trees sorting. The main research works are as follows:1. After comprehensive analysis of the paired comparison algorithms, the high time complexity of the Ortho-reverse algorithm is improved in this paper. In this paper, the adaptive reference ranges of the initial sequence for each participant, ranges from single ranked in the original algorithm to all laboratory submission, effectively avoid the "reverse" phenomenon, which greatly reduces the number of experiments, and the time complexity, so that the performance of the algorithm is superior. Through the simulation with a large number of analog data, this paper validates that the time complexity of the algorithm has been improved in the context of assuring the quality of test results.2. Combined with the characteristics of constructions of the red-black tree, respectively from the MOS evaluation model analysis, paired comparison algorithm improvement, video samples matched information targeting three of the subset of features and analysis, a paired comparison algorithm based on red-black tree sorting is proposed p, that is also called RBTS. Taking advantage of the construction features of the red-black tree, cleverly using the classifications in the construction process, this paper substantially reduces the time complexity, so as to shorten the testing time, improves the operability of the paired comparison algorithm. The time complexity of the algorithm is(?), is a more efficient algorithm. By comparison with other algorithms through simulation experiments, it shows that the results of the algorithm in this paper have smaller deviation by the real values, with better accuracy and stability.