Click here for full text:
Transform Domain Robust Filtering
Nachlieli, Hila; Staelin, Carl; Fischer, Mani; Shaked, Doron; Keshet, Renato; Kisilev, Pavel
Keyword(s): denoising; image processing; over complete transform
Abstract: Image quality is one of the main requirements for imaging devices, such as cameras, scanners, printers, etc. In particular, one is interested in obtaining sharp images that contain as little noise as possible. While the level of noise in captured images is determined mainly by the quality of the hardware (e.g., the CCD in cameras), it can usually be reduced by image processing tools. This report presents an image processing scheme for reducing noise in images. The proposed scheme consists of using robust- estimation filtering methodology in the context of Donoho's state-of-the-art translation-invariant soft- threshold denoising scheme. Specifically, we replace the soft-threshold function in Donoho's scheme with a robust-estimation look-up table, which reduces the blurring common to the approach. Instead of a Wavelet transform (which is usually used in this context), we prefer to use a block-DCT, since it is efficiently implemented in most image capture devices. Extensions and applications of this algorithm for video processing are also proposed. This approach could be seen as part of a second generation of robust- estimation filters, available today in several HP imaging products.
Back to Index