An information preserving data compression for X-ray CT images using MMSE denoising
Kazuki Ogo, Motohiro Tabuchi, Nobumoto Yamane · 2013
Image restoration methods based on a universal Gaussian mixture model (UNI-GMM) may realize minimum mean square error, under locally stationary assumption. Because the UNI-GMM appeared in the literatures observes the model in fixed size square blocks for simplicity, it has trade-off relation, i.e. large blocks become inconsistent to stationary assumption and small blocks diminish noise reduction performance. Arbitrary shaped observation block is known effective in this problem. In the case of UNI-GMM, multi-size observation block is under study to improve consistency of the locally stationary assumption. In this paper, this method is applied for information preserving X-ray CT image compression, in order to improve not only image quality but also compression rate in the diagnostic imaging systems, e.g. PACS system.