Information Sampling for Optimal Image Data Selection

This paper presents a statistical method termed Information Sampling for selecting the most relevant data from an a priori set of images. These data could be a single pixel or a number scattered throughout an image. The main problem addressed is how to determine which image data points contain the most relevant information. As distinct from other techniques, we utilize the inherent information contained within the image set. We show that the most relevant data points, i.e. the most discriminatory, are those which allow reconstruction of a given image with the smallest amount of error. We apply Information Sampling, yielding the most informative data and subsequently rank this data from most to least discriminatory. We show how Information Sampling can be applied to determine the qualitative position of a mobile robot in an indoor environment, using only the highest ranking data.