Stripe Removal from Hyperspectral Food Images acquired by Handheld Camera using ℓ2,1 Norm Minimization and SSTV Regularization

Abstract

Hyperspectral imaging offers the capacity to quickly and noninvasively monitor a food product’s physical, chemical and morphological properties. Specim IQ is a handheld push broom camera with basic data handling and data analysis capabilities within the camera software. However, the recordings of the Specim IQ camera showed a line pattern (stripes) that was evident in all images. Stripes significantly reduce the visual quality of the images and lower the results of further processing. Hence an efficient destriping model is developed, which specifically addresses this issue. The proposed model uses a spatial gradient term to analyze the directional characteristics and group sparsity to describe the structural characteristics of the stripe component. In addition to this, a spatial spectral total variation regularization is used to ensure piecewise smoothness in the spatial and spectral domains and to remove Gaussian noise. The ensuing optimisation problem is solved using the alternating direction method of multipliers (ADMM). The proposed method is tested in real stripe noise environments, and the findings demonstrate that it outperforms some of the best approaches in terms of visual quality and quantitative evaluations. When compared with the other approaches, the proposed method attained the highest noise reduction (NR) and lowest mean relative deviation (MRD) values (NR=1.67, MRD=1.02%).