통계적 방법을 이용한 2-G Bayer CFA Pattern에서의 보간 방향 결정 방법

초록

Most of digital still cameras employ CFA(Color Filter Array), an array of red, green and blue color filters, attached CCD(Charge Coupled Device) sensor to capture color images. Each pixel of the image obtained by the CFA CCD based digital still camera has only one primary color value. Thus, to form a complete color image, the remaining two primary color values should be interpolated using adjacent pixel values. This process is often called as the missing pixel interpolation. Various kinds of missing pixel interpolation methods have been devised and the results are still unsatisfactory. This paper proposes a statistical color interpolation technique for the 2G-Bayer pattern that is the most popular CFA pattern. The proposed method utilizes the statistical discriminant techniques to determine the direction for interpolation. The coefficients of the quadratic decision function are determined using the training sample images. At each pixel, the direction for interpolation is chosen based on the calculated value of the quadratic decision function. The proposed method is compared with Lu’s method which, of course, is for 2-G Bayer CFA pattern and known to be the most accurate one.