Lack-of-Fit Detection Using the Run-Distribution Test

Abstract

In this paper, we are concerned with the problem of deciding whether a fitted model accurately describes the data to which it has been fitted. We present an effective method of testing the lack-of-fit of a parametric model to data, with applications to computer vision. Our test is important to the computer vision community in two ways: We assume a broad enough class of distributions as to be essentially distribution independent . The test requires no knowledge of the sensor noise level . We present results of experiments that compare the test with the standard χ^2 statistic. The experiments are designed to represent typical vision tasks, namely feature tracking and segmentation. We show that our test is more sensitive than the χ^2 unless the noise level is perfectly known.