Statistical properties of the spatiotemporal center-weighted median (CWM) filter for image sequences are investigated. It is statistically shown that the CWM filter preserves image structures under motion at the expense of noise suppression. To improve the CWM filter, a filter which can be effectively used in image sequence processing, the adaptive directional CWM filter (ADCWM) is proposed. This filter utilizes a multistage filtering structure based on adaptive symmetric order statistic (ASOS) operators which produce a pair of order statistics symmetric about the median. The ASOS's are selected by using adaptive parameters adjusted by local image statistics. It is shown experimentally that the proposed filter can preserve image structures while attenuating noise without the use of motion estimation.
|Number of pages
|IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing
|Published - 1993 Aug
Bibliographical noteFunding Information:
Various motion-compensated temporal filtering techniques have been proposed to overcome this problem -. Some of these techniques utilize a combined segmentation and motion detection algorithm to segment the images into moving and non-moving areas, applying a temporal filter only in the nonmoving regions. These methods can preserve image structures under motion, but cannot reduce noise in moving regions. Although noise in moving areas is perceptually masked to some extent by the motion, it will be visible in jlowly moving areas. Other techniques estimate the motion path of a pixel and the temporal filtering is performed over this trajectory so that the moving objects are not distorted. However, the amount of i o k suppression which can be attained with the 1D temporal filter is q.:, 4ted. It is possible to increase noise suppression by utilizing Tiporal windows. However, this leads to a computationally Manuscript received November 7, 1991; revised manuscript received November 19, 1992, and February 22, 1993. This work was supported by the Korea Scisnce and Technology Foundation. This paper was recommended by Associate Editor G. S. Moschytz. S.-J. KO is with the Department of Electronic Engineering, Korea University, Seoul 136-701, Korea. T,M, Forest is with the Department of Electrical and Computer E,ngineering, University of Michigan, Dearbom, MI 48128. iEEE Log Number 9209017.
ASJC Scopus subject areas
- Signal Processing
- Electrical and Electronic Engineering