Adaptive Demosaicing Algorithms Based On Bayer Color Filter Array

Demosaicing is an essential part of color imaging process in digital cameras, video cameras and other imaging equipment. With the wide application of the digital imaging equipment in monitoring, consumption and other field, the research associated with the theory and technology of demosaicing is continuing. From the early simple linear demosaicing to now relatively mature adaptive demosaicing method, each method has its own research focus. Some algorithms are in pursuit of high visual quality of the reconstructive image. Mostly it is used directly dealing with Bayer image by computer. Some algorithms are in pursuit of be suitable for hardware implementation. When the image quality meets certain requirements, the algorithm needs to reduce the processing time as possible. However, a lot of good demosaicing algorithms already exist. Their key technology is still controlled by well-known foreign sensor manufacturer. Therefore, the demosaicing algorithm is usually kept confidential or subject to patent restrictions. That is why doing research on the adaptive demosaicing algorithm has important theoretical significance and application value.First, a low complexity demosaicing algorithm based on direction flag is proposed. Define this flag as binary flag, which judging by adjacent green component. According to 0 or 1 of the flag, give current pixel different weight coefficient and different color component, so as to reconstruct the green component. And the PSNR of green component is up to 37.7dB. A common problem in demosaicing is that the visible color artifacts in high-frequency regions are caused primarily by aliasing in red and blue component. Fortunately, there is usually high inter-channel correlation in high-frequency among red, green and blue channels. Thus, reconstruct red and blue component by adding the high-frequency of green component to other color components. And the PSNR of red and blue component is 36.8dB. The experimental results show that our proposed demosaicing algorithm outperforms other similar algorithms both in PSNR and computation cost.It also presents a high-quality demosaicing algorithm based on homogeneity projection. In order to make Sobel operator suitable for mosaic image to extract more accurate gradient information, it embed the rules of the bilinear demosaicing into the typical Sobel operate. Analyzing the correlation between spatial and spectral, define a spatial-spectral correlation value, which can be used as characteristic of Bayer image's homogeneity. By calculating the homogeneity projection value, choose the best neighborhood, which contains of the most relevant neighboring pixels for current pixel. According to this neighboring pixels and gradient information, reconstruct the three components. The PSNR of the full color reconstructive image is up to 40.68dB, and the human visual different valueΔE a*b is 1.2598. The experiment results show that our proposed demosaicing algorithm has the best PSNR value andΔE a*b value when compared to four similar algorithms.In this paper, two adaptive demosaicing algorithms have been proposed. One is easy to implement. It can be applied in real-time situation or the hardware's cost-limited situation, such as wireless video transmission and remote network monitoring. The other one is difficulty for hardware implement. It can be used in the field demanding high-quality image, such as satellite, biology, medicine etc special image process.