Hui-Jung Lee

Abstract

Images are usually corrupted by noise which comes from various sources: noise in the recording media (e.g. film grain noise), and noise introduced in the transmission channel. Noise degrades the visual quality of images and obscures the detail information in the images. One of the major sources of noise for images recorded on films is film grain noise. An orthonormal expansion algorithm for digital image noise suppression is implemented. The objective is to preserve as much sharpness and produce as few artifacts in the processed image as possible. The method sections an image into non-overlapping blocks. Each block is treated as a matrix which is decomposed as a sum of outer products of its singular vectors. The coefficient of each outer product is modified by a scaling function and the matrix is reconstructed. The resulting image shows a reduction of noise. The two major problems in the method are: 1. the blocking artifacts due to the sectioned processing, and, 2. the trade-off between the suppression of noise and the loss of sharpness. By separating the image into the low frequency and the high frequency components and processing only the latter component, the method is able to reduce the blocking artifacts to an invisible level. To obtain the optimal trade-off between the suppression of noise and the loss of sharpness, systematic variations of the coefficient scaling function were used to process the image. The best choice of the scaling function is found to be [ 1 - (σi / ai ) 3 ] which is a little different from the least-square-error estimate, [ 1 - (σi / ai ) 2 ].

Library of Congress Subject Headings

Image processing--Digital techniques; Matrices--Data processing; Images, Photographic--Data processing

Publication Date

1989

Document Type

Thesis

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Kitchen, Andrew

Advisor/Committee Member

Anderson, Peter

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TA1632 .L434 1989

Campus

RIT – Main Campus

Download

Share

COinS