Demosaic Algorithm Based on Color Difference and Regional Gradient Weight
Xiaojun Zhang, Zhaoyang Ju, Xiaobing Huang, Bincheng Xie · 2024
In hardware image process, demosaicing is the most important step. Because each pixel in bayer image only contains one color channel in RGB, bayer image must be interpolated and reconstructed before it can be restored to full color image. Involving interpolation means that compared with the original image, there will be color cast, edge blur and other phenomena. In order to solve the above problems, this paper proposes a demosaicing algorithm considering the color correlation and color change relativity of adjacent pixels. Considering the overall horizontal and vertical image gradients in the adjacent areas of operating pixels, the ratio of the two directional gradients is calculated to give directional weights to the operating pixels, and the weights are combined with the principle of the same color difference to realize interpolation. This paper compares other demosaicing methods to demonstrate the advantages of proposed demosaicing method. Finally, the demosaicing image performs well in peak signal-to-noise ratio, and its structural similarity is better than other methods.