Motion Compensated Spatial-Temporal Reduction of Film Grain Noise in the Wavelet Domain

The worldwide spread of new display technologies and the advent of HDTV increase the demand for high-resolution image sources. Film is particularly suitable to fit these requirements and is most commonly used for high quality film and video productions. But film and video sequences are often visually distorted by noise resulting from the process of image acquisition, storage and transmission. Film grain noise may reduce the visual quality and becomes visible especially for light-sensitive films, short viewing distances or large screens. The reasons for film grain noise are versatile, e.g. film material, light exposure and development. It differs from other types of noise due to its origin and particular properties. Film grain noise is non-pixel-based and depends on the image content. Therefore common mathematical noise models can only be used restrictedly to describe film grain noise. We propose a new adaptive noise reduction scheme for film grain noise. The algorithm combines spatial and wavelet domain techniques and performs temporal and spatial filtering. The proposed algorithm utilizes five successive images for each noise-reduced film image. In order to get a temporal and spatial signal-noise separation, the algorithm uses bidirectional motion compensation with adaptive image adjustment. Next a three-dimensional wavelet transform is computed and the detail bands are adaptively modified. A motion adaptive three-dimensional median is used in spatial domain for further noise reduction and suppression of local color as well as brightness variations.

Published

2005-11

Content type

Original Research

DOI

10.5594/M00404

ISBN

978-1-61482-942-3