Image Quality Assessment Metrics for Hyperspectral Image Compression Algorithms

When it comes to hyperspectral image-based object detection, image quality is one of the most significant variables to consider. Evaluation of hyperspectral image quality based on the ground truth is necessary for authenticity. In actuality, however, it might be quite challenging to locate the truth of the matter. The evaluation of image quality is carried out using full reference measures, such as MSE (Mean Square Error) and PSNR (Peak Signal to Noise Ratio). In contrast to MSE and PSNR, two additional full reference metrics known as SSIM (Structured Similarity Indexing Method) and FSIM (Feature Similarity Indexing Method) have been developed in recent years with the intention of comparing the structural and feature similarity measures between reconstructed and original HS images on the basis of perception. The comparison of several image quality criteria is the major focus of this article in order to provide a complete picture.However, when viewed from the standpoint of representation, SSIM and FSIM are normalized, although MSE and PSNR are not. When viewed from the viewpoint of semantics, MSE and PSNR only give the absolute error, whereas SSIM and PSNR give the error based on perception as well as error based on saliency. Therefore, the SSIM and FSIM can be considered to be more easily comprehended than the MSE and PSNR. Some other image quality metrics for the hyperspectral image compression such as Bjontegaard Delta PSNR, Maximum Difference, Average Difference, Normalized Quality Index, Normalized Cross-correlation, Structural Content, Normalized Absolute Error and Structural dissimilarity can also use for the analysis.