Feature selection for blind image quality assessment

Abstract

In recent years, reference-less image quality assessment (NR-IQA) has become an important but diftlcult research probe me in the era of digital image processing. No reference IQA purpose is to fond a computational model to predict the subjective quality from tf1e distorted image itself without any reference image. Even though to develop several NR-IQA techniques the excessive work has been done but the issue of NR-IQA is still a challenging task ‘because reference images are usually unavailable in real world scenarios. Human eyes are considered as the standard to recognize visual content that conducts visual analysis for multiple purposes. However, opposed to the large quantity of visual data makes it difficult to manually analyze the image quality by using the human observer. TQA using human observers consumes a large amount of time and it is a tedious task. Therefore computational models that use the objective model to predict the quality score of an image have gained importance. The performance of NR-IQA techniques, objective matrix depends on the quality of features. In literature, many techniques have been proposed to extract features from images in different domains but the extracted features may tend to contain redundant features that reduce the performance of IQA techniques to predict the quality score~,. The proposed methodology presents seventeen new feature selection algorithms specifically designed for NR-IQA, which are based on Perceptually Weighted Rank Correlation (PWRC), Spearman rank-ordered correlation constant (SROCC), linear correlation constant (LCC), Kendall correlation constant (KCC), root mean squared error (RMSE) and Mean Absolute Error (MAE). The proposed feature selection algorithms are applied to the extracted features of existing NR-IQA techniques. Support vector regression (SVR) is then applied to selected fee trues to predict the image quality ~ core . The seventeen newly proposed feature selection algorithms are veal rated using eight different NR-IQA techniques over four commonly used image quality assessment clambakes. Ex~ perimental result its show that the proposed feature selection algorithms not only ~educe the number of features but also improve the performance of NR-IQA techniques. Moreover, feature selection algorithms based on PWRC and its combination with SROCC, LCC, KCC, MAE, and RMSE perform better in comparison to other proposed algorithms.