Denoising Poisson Noise In Low Photon Counts X-ray Spectral Images With Weighted Bootstrap Dictionary Learning Method

Recently,with the development of spectral imaging techniques,X-ray spectral image has been able to provide the nanoscale spectral resolution,which can show the specific spatial distribution of different trace element in biomedical samples.So more and more researchers in biomedical field have been paying attention to the X-ray spectral.However,in the photon counting detection process of characteristic X-ray,the detected data will inevitably be contaminated by Poisson noise.On the one hand,the content of trace elements in biological samples is very low,the intensity of excited characteristic X-ray and the detected photon counts on the detectors becomes very low.Thus it causes the low photon counting problem.On the other hand,the amount of obtained samples for detecting the distribution of specific trace elements is far from enough.Therefore,the X-ray spectral images with low photon counts problem also have the small sample problems.Furthermore,because of the small sample problem,the self-similarity inside images is insufficient,some state-of-the-art image denoising methods are not able to obtain the good denoising performance.In this work,we denoise the X-ray spectral with low photon counts problem using the bootstrap resampling method and principal component analysis method under the dictionary learning framework.Firstly,we use the patch-ordered operations to sort the extracted image patches and resample the different blocks of spectral images using bootstrap resample method to obtain more data matrix of image samples for training the initial dictionary.Then,we select the data matrices with top largest similarity weights as the initial dictionary.Finally,we update the atoms of dictionary by extracting the principle component of the PCA decomposition of data matrices to obtain the final denoised image.We use some experimental data to compare the different denoising performances of our proposed algorithm and other state-of-the-art image denoising methods.The experimental results show that our method offers satisfying results for X-ray spectral images with the small image size.