Automatic identification of rank correlation between image sequences

Correlation is one of the most common statistical tools and one of the most basic methods for discovery from data. Over the past century, correlation has led to an extremely high number of discoveries in virtually all fields of science and is currently one of the basic building blocks that enable scientific research. In most cases, correlation is detected between numerical variables using tools such as Pearson or Spearman correlation. This paper describes a method that can identify rank correlation between two sequences of images. That is, instead of identifying a rank correlation between two numerical variables, the method identifies rank correlations between two variables such that each variable is an image. Experimental results show that the method is able to detect rank correlations between sequences of synthetic images, as well as sequences of natural images. The method can provide useful information for profiling visual data, and example discoveries that can be made with the method are discussed. Given the prevalence of correlation in data-driven discovery, combined with the increasing size and availability of databases of visual data, identifying rank correlations in image data can provide useful insights and new knowledge from the data. The method is nonparametric and can be applied to various types of image data.