Color reproduction pipeline for an RGBW CFA sensor

Abstract

Many types of RGBW color filter array (CFA) have been proposed for various purposes. Most studies utilize white pixel intensity for improving the signal-to-noise ratio of the image and demosaicking the image. This dissertation proves for the first time that the accuracy of white balance can be improved using the white pixel. For detecting gray pixels, which is a key clue for the illuminant estimation, the conventional methods have difficulty in detecting reliable gray pixels from uniform regions and boundaries of different color surfaces. The proposed method can accurately and quickly detect gray pixels even in uniform color regions by using the additional white pixel intensity. In order to reduce computational complexity and eliminate parameter dependence, an adaptive thresholding method is proposed that can adaptively select reliable gray pixels using the characteristics of the Skellam distribution. The image processing processor including the proposed methods is implemented in hardware. Furthermore, an adaptive white-preserving color correction is proposed that adjusts the color correction matrix according to the grayness of each pixel calculated from the detected gray pixels. According to the experimental results of the dataset captured with the RGBW CFA OPA sensor, the proposed gray pixel detection method was 22.1% more accurate and 30.4 times faster than the conventional methods. The proposed adaptive color correction method corrects achromatic color 37.4% more accurately than the conventional method. Therefore, the proposed color reproduction pipeline for the RGBW CFA sensor is expected to be used to obtain images with high color accuracy in mobile devices, and furthermore, it is expected to be used to improve the accuracy of recognition algorithms used in robotics, advanced driver assistance systems, and CCTV.