An Ensemble Learning Based Adaptive Algorithm for Capsule Endoscope Image Deblocking

摘要

In capsule endoscope system, discrete cosine transform (DCT) compression on medical image brings annoying mosaic effect. Previous methods focus more on perfect visual smoothness without considering detail preservation. To address this issue, we integrate ensemble learning model into projection onto convex set (POCS) method. Both structure features and severity of blocking artifacts are evaluated by learning model, and the learning results are used to adaptively modify parameters of constraint convex sets. Finally, we obtain a visually smooth diagnosis image with good detail preservation and an average peak signal to noise radio (PSNR) of 42.90.