Correlation of inter-channel high-frequency components in demosaicking weak correlated CFA data

Abstract

A novel method for obtaining local high-frequency components of the missing colour channel in weak correlated colour filter array (CFA) data, by calculating the local inter-channel high-frequency correlation coefficients, is proposed. The contrast of experimental results demonstrates that the high-frequency components achieved by the proposed method are closer to the original image than those from state-of-theart approaches.Introduction: In a single chip sensor array, a colour filter array (CFA) is placed between the lance and the sensor for image acquisition. The CFA image is essentially a greyscale mosaic-pattern image as shown in Fig.1. The process for restoring the RGB full colour image from the CFA data is called colour interpolation or colour demosaicking.Gunturk et al. and Chen et al. have demonstrated that the highfrequency component of the green plane and that of the red/blue plane are highly correlated, the correlation values ranging from 0.98 to 1 [1, 2]. Most of the existing demosaicking algorithms replace the high-frequency components of the missing channel with those of the alternative pixel channel [1-4]. However, Zhang et al. and Chang and Tan have found that the assumption of high-frequency correlation does not hold in areas of highly saturated colours (weak correlated image) [5,6]. Zhang supplements an existing colour demosaicking algorithm, which exploits the same inter-channel high-frequency correlation as methods in [1][2][3][4], by combining its results with linear minimum mean-square estimation and the support vector regression. Chang utilises the spatial and spectral correlations to interpolate the CFA image. Their methods perform better than previous algorithms, especially when restoring the weak correlated CFA image.This Letter focuses on the local relationship of inter-channel highfrequency components, instead of looking at it as one global problem as previous methods, when interpolating the weak correlated CFA image. The contribution of this Letter is acquiring the locally varied high-frequency components of the weak correlated CFA image, according to the local inter-channel high-frequency correlation coefficients in the wavelet domain. Experimental results show that the proposed method outperforms or performs close to the state-of-the-art approaches.

Demosaicing scheme based on augmented inter‐channel correlation considering colour saturation

“…However, high-frequency loss occurs during full colour reconstruction. Since the development of the Bayer pattern, various demosaicing algorithms have been studied in order to minimise the degradation of highfrequency component [2,3]. The demosaicing algorithm proposed in [2] allows local high-frequency components in a weakly inter-channel correlated pattern to be obtained.…”

“…Since the development of the Bayer pattern, various demosaicing algorithms have been studied in order to minimise the degradation of highfrequency component [2,3]. The demosaicing algorithm proposed in [2] allows local high-frequency components in a weakly inter-channel correlated pattern to be obtained. Further, the recently developed multiscale gradient (MSG) algorithm [3] constitutes one of the most efficient demosaicing methods, implementing an edge adaptive approach.…”