Blind infrared spectral deconvolution with discrete Radon transform regularization for biomedical applications

Abstract

Infrared spectrum often suffers from the resolution reduction and random noise. This paper proposes a novel blind infrared spectral reconstruction model that integrates total variation constraint and frequency domain transformation. This model aims to achieve an accurate deconvolution model of infrared spectra by making the coefficient distribution of discrete Radon transform (DRT) of overlapping infrared spectra close to high-resolution infrared spectra. Secondly, we use total variation (TV) as a popular effective spectral prior model, which has been applied in regularization based blind deconvolution of infrared spectra because it can preserve small peaks. In this study, the model fully utilizes spatial information from different image regions and proposes an extended split Bregman iteration method to solve the joint minimization problem. Specifically, the DRT coefficient distribution of overlapping infrared spectra should be close to high-resolution infrared spectra. We believe that there are differences between the DRT coefficient distribution of clean spectra and the distribution of degraded infrared spectra. Extensive experimental results have shown that the proposed method outperforms most existing methods in terms of spectral structure quality and quantitative measurement. The high-resolution infrared spectra after deconvolution can be used for biomedical imaging and clinical applications.