Learning Markov random field image prior for pixelation removal of fiber microscopy using sparse coding based on Bayesian framework

We were able to efficiently remove the morphological artifact of the fiber bundle based endo-microscopy and improve the featured patterns within the object image acquired by using non-invasive near infrared optical coherence tomography. Our image reconstruction methodology starts to estimate the original shape from the regions that are directly damaged from the en face image which contains significant image degradation by the pixelation of numerous imaging fiber units. Then we have iteratively extended the neighbor areas from the initial status so that we can successfully estimate the original shape of the missing pattern.