Destriping of Landsat MSS Images by Filtering Techniques

Abstract

The removal of striping noise encountered in the Landsat Multispectral Scanner (MSS) images can be generally done by using frequency filtering techniques. Frequency domain filtering has, however, several problems, such as storage limitation of data required for fast Fourier transforms, ringing artifacts appearing at high-intensity discontinuities, and edge effects between adjacent filtered data sets. One way for circumventing the above difficulties is to design a spatial filter to convolve with the images. Because it is known that the striping always appears at frequencies of 1/6, 1/3, and 1/2 cycles per line, it is possible to design a simple one-dimensional spatial filter to take advantage of this a priori knowledge to cope with the above problems. The desired filter is the type of finite impulse response which can be designed by a linear programming and Remez's exchange algorithm coupled with an adaptive technique. In addition, a four-step spatial filtering technique with an appropriate adaptive approach is also presented which may be particularly useful for geometrically rectified MSS images.