Quality Metric for Multispectral Image Compression

Abstract

A new metric, blockwise distortion measure for multispectral images (BDMM), is proposed to evaluate the quality of compressed multispectral images. The metric is based on blockwise distortion measures which are computed between the original multispectral image and the compressed/reconstructed multispectral image. The computed measures between the two images are adapted to a visual scale which is based on the mean opinion score (MOS). The visual scale is obtained from visual comparison of spectral bands of multispectral images. Two methods were used for matching the MOS-values with the measures computed from the images: the nonlinear weighting using the neural network gave better matching than the linear model by optimization. The measure with empirically defined parameter values is given. The results from the new metric are compared to mean-square-error (MSE) based measures, to blockwise distortion measure, and to picture quality scale for grayscale images. Multiwavelets, the principal component analysis, and JPEG are used in the compression of the multispectral images. The metric can be used with different multispectral images and with different compression methods.