1 December 1996 New feature-preserving filter algorithm based on a priori knowledge of pixel types
Author Affiliations +
Abstract
The concept and algorithmic details of a new corrupted-pixelidentification- (CPI)-based estimation filter are presented. The approach is by transforming a noisy subimage centered on a corrupted pixel into its discrete cosine transform (DCT) domain, and approximating the transformed subimage by its DC (average) coefficient only, an estimation of the noise distribution is made by combining the knowledge of the number of corrupted pixels in the subimage and the pixel intensity of the noise term. This enables the DC coefficient of the restored image in the DCT domain to be determined, and from this, the restored pixel intensity can be calculated by an inverse DCT. The whole restored image can be obtained after all the corrupted pixels are exhausted. From an extensive performance evaluation, it was found that the new algorithm has a number of desirable characteristics. First, the CPI-based estimation algorithm performs extremely well when heavily degraded images are concerned. Second, the CPI-based estimation algorithm has acceptable featurepreserving properties, far better than the conventional median filter. Third, the new algorithm can be applied iteratively to the same noisy image. Fourth, the computing speed of the CPI-based estimation algorithm is almost three times faster than the conventional median filter, and 1.6 times faster than the original CPI algorithm, making it the fastest algorithm in this class so far.
Andrew H. S. Lai, Andrew H. S. Lai, Nelson Hon Ching Yung, Nelson Hon Ching Yung, } "New feature-preserving filter algorithm based on a priori knowledge of pixel types," Optical Engineering 35(12), (1 December 1996). https://doi.org/10.1117/1.601087 . Submission:
JOURNAL ARTICLE
14 PAGES


SHARE
RELATED CONTENT

Image scaling with aliasing cancellation
Proceedings of SPIE (January 24 2011)
Quantitative evaluation of image enhancement algorithms
Proceedings of SPIE (June 01 1991)

Back to Top