Use of color image correlation in the retrieval of gastric surface topography by endoscopic stereopair matching

Abstract

We propose finding an optimal generalized color plane to carry out digital color image correlation. We seek the optimization of the correlation peak's sharpness and, for the purpose of analysis, we make some assumptions regarding the statistical properties of images to be considered. The analysis leads to a simple solution involving a projection onto the Karhunen-Loeve vectors corresponding to the smallest eigenvalues of the interprimary colors covariance matrix. We use our results in the retrieval of the gastric mucosal surface structure from stereopairs taken through an endoscope. Study of the statistics of the endoscope images shows that the assumptions used earlier can be applied to this type of image. We present some results of stereopair matching with color image correlation. We conclude by proposing a pseudo-3-D surface reconstruction of the gastric surface by mapping the endoscopic image onto the retrieved elevation information.