Lanczos Resampling for the Digital Processing of Remotely Sensed Images

Abstract

This paper presents theoretical and practical application of a relatively unknown and rare image resampling technique called Lanczos resampling. Application of this method on satellite remote sensing images is considered. Image resampling is the mathematical technique used to create a new version of the image with a different width and/or height in pixels. Interpolation is the process of determining the values of a function at positions lying between its samples. Sampling the interpolated image is equivalent to interpolating the image with a sampled interpolating function. Image registration is the process of overlaying two or more images of the same scene taken at different times, from different viewpoints, and/or by different sensors. It geometrically aligns two images: the reference and sensed images. In the interaction between interpolation and sampling processes, aliases occur on some occasions. Majority of the registration methods consist of the steps like feature detection, feature matching, transform model estimation and image resampling and transformation. The proprietary softwares that are commercially available for image processing that are capable of doing image registration do not provide us with performance metrics for assessing the resampling methods used. Lanczos resampling method has not been used in the digital processing of remotely sensed satellite images by any of the open source and the proprietary software packages that are available until now. In this paper, we have applied performance metrics (on satellite images) for analyzing the performance of Lanczos resampling method. Comparison of Lanczos resampling method with other resampling methods, such as nearest neighborhood resampling, and sinc resampling, is done based on the metrics pertaining to entropy, mean relative error, and time. We propose that Lanczos resampling method to be a good method from qualitative and quantitative point of view when compared to the other two resampling methods. Also, it proves to be an optimal method for image resampling in the arena of remote sensing when compared to the other methods used. This, we hope, will enhance the understanding of the classified images’ characteristics in a quantitative manner.