The Inverse Compositional Algorithm for Parametric Registration

Abstract

We present an implementation of the inverse compositional algorithm for parametric motion estimation. It computes a global motion between two images using a non-linear least square technique. Our implementation allows computing several types of planar transformations, such as translations, similarities, affinities or homographies. The algorithm is iterative so it typically yields solutions with high accuracy. The use of robust error functions, different from the L 2 norm, improves the stability of the method under the presence of noise and occlusions, and allows it to detect the predominant motion, even if there are several types of displacements. The method works with multi-channel images and makes use of a coarse-to-fine strategy for dealing with large displacements. Source CodeThe reviewed source code and documentation for this algorithm are available from the web page of this article 1 . Compilation and usage instruction are included in the README.txt file of the archive.

Classic and inverse compositional Gauss‐Newton in global DIC

“…From this point of view, the most mature method in subset-DIC is certainly the inverse compositional Gauss-Newton (ICGN). [12][13][14][15][16][17] The elegant idea behind is to perform a specific linearization of the minimization problem in such a way that the use of a Gauss-Newton naturally leads to a constant operator. For global approaches and, thus, for FE-DIC, the linear operator is made constant due to the following trick: at iteration k, the gradient ∇g • k (that is of the deformed image g advected by the current approximation of the mapping k ) is replaced by the gradient ∇f (ie, of the reference image f ).…”

“…The family of compositional Gauss-Newton algorithms was first introduced in image processing by Baker and Matthews. 12 Many variants (inverse, forward, compositional, additive, etc) were then developed 13,24 and applied to subset-DIC 14,16,17,24 and subset-DVC. 15 The aim is still to develop a fast and robust solver with a constant operator.…”

“…From this point of view, the most mature method in subset‐DIC is certainly the inverse compositional Gauss‐Newton (ICGN) . The elegant idea behind is to perform a specific linearization of the minimization problem in such a way that the use of a Gauss‐Newton naturally leads to a constant operator.…”

Comparison of Motion Smoothing Strategies for Video Stabilization using Parametric Models

“…In general, the homographies are representatives of the background motion, where the number of inliers is usually bigger. On the other hand, we use a direct method [49] implementing the inverse compositional algorithm [4,3], which is an improvement of the Lucas-Kanade Algorithm [42]. It can compute any of the transformations in Table 1.…”

“…In the code and online demo, we use a classic feature-based technique [44], which relies on SIFT features [41], and a direct method [49], which implements the inverse compositional algorithm [4,3]. Direct methods are usually faster but more sensitive to brightness changes and outliers.…”