Improved physical-like virtual color filter for color separation of White-RGB CMOS sensor

White-RGB (WRGB) color filter array (CFA)-based CMOS sensor is a new and popular technique to improve the image quality directly with a higher sensitivity White (W) filter. Considering there is no color information in the pixel assigned to W, color separation from W raw data is necessary. We propose a color separation method based on a novel physicallike virtual color filters model for the WRGB CFA-based imaging system. The proposed algorithm can reproduce full color signal from W. In the filtering stage, the proposed algorithm provides three channel filter modes to imitate the physical color filter based on the wavelength responses. To reduce the deviation of the spectrum, a supervised learning technique is utilized to obtain the parameters. The Datacolor SpyderCheckr-based experimental results demonstrate that the proposed algorithm provides high accuracy color without the color artifact caused by the error of virtual filter, and the average improvement compared with the Gaussian ideal filter model is almost 90%.

I. Introduction

Most of digital cameras use a single-chip image sensor, which is spectrally shifted alternating patterns of several color filters applied to each pixel location called color filter array (CFA) [1]. Many different CFA patterns have been proposed over the years. Most of them are comprised of pure RGB channels, such as Bayer pattern shown in Fig. 1(a). Half of the pixels in the Bayer CFA are assigned to G channel, and the rest of the pixels are equally shared by Rand B channels. Considering Green is more sensitive to human eyes, More G filters are assigned in traditional CFA patterns. However, G filter is not sensitive enough especially in low-light environment. Therefore white (W) filter is used instead of half of G. This novel CFA pattern is named as White-RGB (WRGB) CFA as Fig. 1(b).