Prediction of Quality in DCT-Based Lossy Compression of Noisy Remote Sensing Images

Abstract : This paper considers specific aspects of lossy compression of noisy remote sensing images. A method based on discrete cosine transform (DCT) in 32x32 pixels blocks is analyzed. Characteristics of noise assumed additive (in original data or after proper variance stabilizing transform), spatially uncorrelated and Gaussian are assumed a priori known. They are taken into consideration in setting quantization step (QS) that is supposed proportional to noise standard deviation. It is demonstrated that statistics of DCT coefficients determined in 8x8 pixel blocks can be employed for prediction of peak signal-to-noise ratio (PSNR) and improvement (or reduction) of visual quality metric PSNR-HVS-M. Approximating curves are obtained by their regression into scatter-plots using low order polynomials. Coder performance prediction can be then used for setting its parameter (QS) for providing appropriate quality of compressed image. Applicability of the proposed prediction approach is proven by experiments with real-life images.
Liste complète des métadonnées
Contributor : Laurent Jonchère <>
Submitted on : Tuesday, October 24, 2017 - 9:22:03 AM
Last modification on : Monday, February 11, 2019 - 2:18:49 PM



S. Abramov, V. Lukin, A. Zemliachenko, B. Vozel, K. Chehdi. Prediction of Quality in DCT-Based Lossy Compression of Noisy Remote Sensing Images. 2017 IEEE 37th International Conference On Electronics and Nanotechnology (elnano), Apr 2017, Kyiv, Ukraine. pp.447-450, ⟨10.1109/ELNANO.2017.7939794⟩. ⟨hal-01622075⟩



Record views