Missing Data Interpolation with Gaussian Pyramids

Satyakee Sen

ssen@sep.stanford.edu

ABSTRACT

I describe a technique for interpolation of missing data in which local operators of many scales but identical shape serve as basis functions. A data structure known as the Gaussian pyramid is developed to represent image information at different scales. This data structure in essence consists of a series of lowpass filtered versions of the original image stacked up one on top of the other forming a pyramid like structure. I first show how to generate a set of reduced images which stack up to form the Gaussian pyramid structure and then show how we can use this Gaussian pyramid structure to fill in missing data. Several examples of filling in missing data with this algorithm are shown and in most cases the results are comparable with those estimated using a prediction filter approach.