Multidirectional weighted interpolation based generic demosaicking for single-sensor multispectral imaging systems

Abstract

Single-sensor based multispectral imaging systems with the multispectral filter array (MSFA) are the low-cost and portable means of capturing multispectral images (MSIs) having applications in different domains. These systems require an effective multispectral image demosaicking (MSID) method to reconstruct the full MSI from the raw image captured using single sensor. The MSFA patterns are also crucial for selecting the number of bands in the MSI and for the efficacy of MSID methods, and binary tree-based MSFAs (BT_MSFAs) are preferred and widely used. Highlighting strong similarity in the pixel arrangements of bands with similar probability of appearance in the BT_MSFAs, we propose a new generic MSID method based on the local multidirectional gradients and a weighted combination of spectral differences to find the missing pixel values of the bands. Experimental results confirm that the proposed MSID consistently outperforms the existing generic MSID methods in terms of various subjective and objective image quality metrics on the three benchmark MSI datasets.