Enhancing terahertz images corrupted by compact impulse noise: feasibility and practical recommendations

The subject matter of this article is the process of enhancing low-quality terahertz images through digital image processing. This study aims to investigate whether the visual quality of terahertz images corrupted by compact impulse noise can be improved while preserving their information value. The tasks to be solved are as follows: 1) to analyze impulse noise in real-life terahertz images obtained by raster scanning systems and to evaluate the typical range of its probabilities and localization features; 2) to create an adequate model for compact impulse noise generation; 3) to establish a relation between the probability of impulse noise, image spatial resolution, filtering settings, and the quality of the output images, and give the corresponding recommendations; and 4) to verify the obtained results on real-life terahertz data. The methods used are: mathematical modelling, numerical simulation, and statistical analysis. The following results were obtained. 1) The shape and localization characteristics of compact impulse noise in terahertz images acquired using a single-detector direct imaging setup were studied, and an appropriate generative statistical model was implemented. 2) The possibility of suppressing stripe-shaped compact impulse noise using a classical median filter and its modification with spatial adaptation was investigated through numerical simulations on a set of test images. 3) It was shown that for images no less than 100×100 pixels large and impulse noise probability up to 0.5, filtering with vertically oriented rectangular windows (9×1 to 15×1 pixels) allows quality improvement up to 45 dB according to PSNR and PSNRHVSM metrics, providing effective noise suppression while preserving object details. 4) The formulated recommendations’ adequacy was verified on real-life terahertz data, and the proposed approach’s workability in the presence of other distortions was confirmed. Conclusions. The scientific novelty of the obtained results is in the confirmation of the possibility of reliable restoration of terahertz images severely distorted by compact impulse noise without their re-acquisition along with providing practical recommendations for filtering settings and the requirements to the input terahertz data.