A Case for Denoising before Demosaicking Color Filter Array Data (original) (raw)

Sung Hee Park	Hyung Suk Kim	Steven Lansel	Manu Parmar	Brian A. Wandell

Asilomar Conference on Signals, Systems, and Computers 2009

Abstract

Denoising algorithms are well developed for grayscale and color images, but not as well for color filter array (CFA) data. Consequently, the common color imaging pipeline demosaics CFA data before denoising. In this paper we explore the noise-related properties of the imaging pipeline that demosaics CFA data before denoising. We then propose and explore a way to transform CFA data to a form that is amenable to existing grayscale and color denoising schemes. Since CFA data are a third as many as demosaicked data, we can expect to reduce processing time and power requirements to about a third of current requirements.

Paper

Result Images

Image #	BLS-GSM	CBM3D
σ = 10	σ = 20	σ = 10	σ = 20
AH	AFD	AH	AFD	AH	AFD	AH	AFD
DM	DN	DM	DN	DM	DN	DM	DN	DM	DN	DM	DN	DM	DN	DM	DN
19	31.7	31.6	32.5	31.7	28.6	28.8	29.5	28.8	32.4	32.2	33.2	32.3	29.8	29.5	30.1	29.5
23	33.7	34.3	34.4	34.2	30.5	31.4	31.7	31.4	35.1	34.8	35.4	34.7	32.0	32.1	32.5	32.0
Avg	31.4	31.6	32.3	31.8	28.4	28.6	29.0	28.6	32.3	31.9	33.0	32.1	29.3	29.1	29.8	29.2

- Demosaicking methods : AH (adaptive homogeneity), AFD (adaptive frequency domain method)
- Pipeline order : DM (demosaick first), DN (denoise first)

[ kodim19, σ = 10 / 20, AH, Before denoising]