Analysis of Weibull Statistic Features Impact on Image Degradation Measurement

Abstract

The traditional concept of quality of service (QoS) which focuses on network performance (e.g. packet loss, throughput, and transmission delay), recently has been grown towards the modern concept of quality of experience (QoE). This reflects all user practice including accessing and service provided. In order to maintain the required QoE, it’s necessary for the service provider to recognize and measure image degradation. This study provides different features in order to assess degraded image quality blindly depending on Weibull statistics. Also, it presents a comparison analysis to give the more performing one. The introduced features are originated from the gist of natural scenes (NS) using Weibull distribution of Log-derivatives. These measuring features were collected through both sharper and rich edging regions of the images. Besides, Weibull features were developed by maximum likelihood estimation (MLE) parameters to improve the quality assessment. LIVE database used to calibrate the proposed features achievement. Experiments prove Weibull statistics the best among popular full-reference peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) methods. Also, they show Weibull features extracted by means of sharper regions are the best when assess the prediction monotonicity. While applying the prediction accuracy evaluation come up with a good performs when taking the improved Weibull features via sharper regions.