An Efficient Approach for No Reference Image Quality Assessment (NR-IQA) Index Using Autoencoder-Based Regression Model (ARM)

In a broad variety of multimedia applications, the accuracy of perceptual quality evaluation is crucial. The purpose of image quality assessment is to mimic human subjective visual perception and to automate the process of image quality inference. In contrast, current NR-IQA systems judge quality simply on the distorted picture, rather than taking into account the influence of the environment on human perception. The no-reference quality evaluation aims to automatically evaluate the perceived quality of the final result when a single picture is created by fusing together a number of band images. This research provides a novel no-reference image quality evaluation approach for satellite image fusion methods. The measure leverages regressor-based autoencoders throughout the assessment process. In the suggested technique, features are derived using the relationship metric and parameters that relate the quality from the pixel of the fused picture by the fine-tuned encoder. Finally, in order to assess and quantify picture quality decline, these components are regressed to quality scores and concatenated. The experimental findings proved that the proposed NR-IQA technique outperforms the existing state of the art on a broad range of NR-IQA datasets, making it suitable for satellite image classification and distortion-type identification tasks.