Development of Iterative Minimum-Maximum Filter for Reducing Impulse Noise from Highly Corrupted Images

Digital images are often corrupted by impulse noise during acquisition or transmission through communication channels. Noisy pixels are characterized by having values that are substantially different from their surroundings. In an environment of fierce noise contamination, infected pixels tend to connect into noise blotches that could give the filtering algorithm an illusion of being part of the original image data. Therefore, many impulses would be difficult to detect, with the consequence of a less chance for proper detection and thus, filtering. Different methods have been introduced in literature to filter images with high noise levels, including non-linear, fuzzy and combined filters. Performance of some typical filters of each category is studied in detail and compared to that of the suggested filter.This study introduces an iterative minimum maximum filter for images highly corrupted with impulse noise, typically in the range 30-80%. Noise detection and filtering are done separately and iteratively, where the impulse detector with a threshold value and the scanning window size, are made proportional to a measure of noise level. Extensive testing, using different types of standard test images, has proved the effectiveness of the proposed filter to give lower Mean Square Error (MSE) of the filtered images. Higher Bit Correct Ratio (BCR) values with better visual quality images have also been recovered compared to other studied filters such as nonfuzzy, fuzzy, and combined filters. This study has verified that a reasonable tradeoff has been achieved between the two aspects of impulse noise suppression and image edges preservation, which are considered as two inherently conflicting requirements. To facilitate use of the proposed filter, the algorithm has been implemented as a stand-alone application, in the form of an attractive graphical user interface.

Preview PDF 549038_FK_2006_27.pdf Download (116kB)

Actions (login required)

View Item