Adaptive Feature Fusion and Kernel-Based Regression Modeling to Improve Blind Image Quality Assessment

In the fields of image processing and computer vision, evaluating blind image quality (BIQA) is still a difficult task. In this paper, a unique BIQA framework is presented that integrates feature extraction, feature selection, and regression using a support vector machine (SVM). Various image characteristics are included in the framework, such as wavelet transform, prewitt and gaussian, log and gaussian, and prewitt, sobel, and gaussian. An SVM regression model is trained using these features to predict the quality ratings of photographs. The proposed model uses the Information Gain attribute approach for feature selection to improve the performance of the regression model and decrease the size of the feature space. Three commonly used benchmark datasets, TID2013, CSIQ, and LIVE, are utilized to assess the performance of the proposed methodology. The study examines how various feature types and feature selection strategies affect the functionality of the framework through thorough experiments. The experimental findings demonstrate that our suggested framework reaches the highest levels of accuracy and robustness. This suggests that it has a lot of potential to improve the accuracy and dependability of BIQA approaches. Additionally, its use is broadened to include image transmission, compression, and restoration. Overall, the results demonstrate our framework's promise and ability to advance studies into image quality assessment.