RGBW array color image reconstruction based on guided filtering and residual interpolation

The Bayer array is a widely used single-chip color filter in digital cameras due to its simple structure and straightforward reconstruction methods. However, as the demand for color imaging in low-light scenarios rises in fields such as reconnaissance, security, mobile imaging, and autonomous driving, the Bayer array, with its narrow spectral range, is unable to capture sufficient spectral information, thereby limiting color imaging capabilities with a high signal-to-noise ratio (SNR). The RGBW array is an extension of the traditional Bayer array, with its channels incorporating near-infrared spectral information. One of the green channels is modified to cover the full visible and near-infrared spectrum, enabling the capture of more spectral data and improving the SNR of color imaging. However, the introduction of near-infrared information through the RGBW filter array distribution, along with its specific layout, necessitates the development of dedicated color image reconstruction algorithms. This paper presents an RGBW array color image reconstruction method based on guided filtering and residual interpolation (GF-RI). An interpolation algorithm is designed, where the spatial domain pixel coverage density progressively increases and pixel-wise optimization is iteratively applied. The high-frequency components of the W channel are utilized to enhance the color image. Experimental results show that, compared to traditional algorithms, GF-RI performs better in reconstructing RGBW mosaic images, producing color images with higher accuracy. Additionally, when applied in real low-light scenarios, GF-RI, in conjunction with a series of image processing algorithms, converts the four-channel RGBW mosaic image into a three-channel RGB color image suitable for human observation. The reconstructed color image exhibits improved detail representation and color restoration, aligning more closely with human visual perception, and enables natural color imaging at the illumination level as low as 10⁻³ lx.