The adaptive SD filter for spatially invariant image sequences

This paper presents a new adaptive form of the simultaneous diagonalization (SD) filter which is both intuitively appealing and mathematically proven to give improved results over the standard SD filter. The SD filter is the linear filter that generates a single new filtered image from an original sequence of spatially invariant (SI) images, such that in the filtered image a desired feature is maximized relative to undesired features and noise. Applications of the SD filter to date have used a constant filter vector, which involves all oft he undesired feature signatures, even in regions where not all features are present. Improved enhancement of the desired feature is obtained with the adaptive filter, which utilizes the results fr om an initial filtering to characterize the feature locations, which in turn indicates the specific form of the adaptive SD filter to use in each region of the image scene.