Gradient based Multispectral Demosaicking Method using Single Sensor Array

Single sensor cameras for RGB color images using color filter array (CFA) in front of the sensor, captures only one color sample among three samples (red, green and blue) at each pixel location. Captured image, from single sensor camera, is called raw CFA image or mosaicked image, and the process of interpolating the missing color samples and reconstructing the full image from the CFA image is called CFA demosaicking. Without increasing cost and size of the camera, the idea of using single sensor and extension of color filter array (CFA) to multispectral filter array (MSFA) enable us to acquire the multispectral image in one shot. However, extension from CFA demosaicking to MSFA demosaicking is not straightforward because of a large number of spectral bands and very sparse sampling of each spectral band in MSFA. In this paper, we propose a generic gradient based multi-spectral image demosaicking method. MSFA patterns are designed using binary tree method. Band sequence arranges in this manner that middle band is set to be with the highest probability of appearance in MSFA pattern. A middle band is interpolated first and using this band, all other spectral bands interpolates with progressive bilinear spectral difference method. We compare our approach with conventional multispectral demosaicking method namely binary tree edge sensing (BTES) method, using CAVE dataset multispectral images. Experimental results show that our proposed approach outperforms BTES with respect to both peak-signal-to-noise-ratio (PSNR) and structure similarity (SSIM).