Matching of large images through coupled decomposition
1. MATCHING OF LARGE IMAGES THROUGH COUPLED DECOMPOSITION
ABSTRACT
In this paper, we address the problem of fastand accurate extraction of points that
correspond to the samelocation (named tie-points) from pairs of large-sized images.First, we
conduct a theoretical analysis of the performance ofthe full-image matching approach,
demonstrating its limitationswhen applied to large images. Subsequently, we introduce a
noveltechnique to impose spatial constraints on the matching processwithout employing
subsampled versions of the reference and thetarget image, which we name coupled image
decomposition. Thistechnique splits images into corresponding subimages througha process that
is theoretically invariant to geometric transformations,additive noise, and global radiometric
differences, aswell as being robust to local changes. After presenting it, wedemonstrate how
coupled image decomposition can be used bothfor image registration and for automatic
estimation of populargeometry. Finally, coupled image decomposition is tested on adata set
consisting of several planetary images of different size,varying from less than one megapixel to
several hundreds ofmegapixels. The reported experimental results, which includescomparison
with full-image matching and state-of-the-arttechniques, demonstrate the substantial
computational costreduction that can be achieved when matching large imagesthrough coupled
decomposition, without at the same timecompromising the overall matching accuracy.