Estimation of horizontal and vertical translations of large images based on columns and rows mean energy matching

Abstract

In this paper mean energy lateral histograms have been developed and used for the translation of frames detection and for the estimation of a shift vector in large image frames. 2-D translation problem is solved by two almost independent 1-D problems based on the mean energy lateral histograms. We formulate a simple optimization for estimating vertical or horizontal translations of the frame. The optimization goal is to find the best match between energies (squared norms) of rows (columns) of the image and the reference frame. The shift estimate is calculated as the minimizer of the least squares, maximum deviation or the sum of absolute deviations criterion. However, the first criterion is preferred, the last two almost always gave the same results when the noise level is low. The proposed method is faster than the FFT based phase-correlation approach. An iterative version of the algorithm is presented which is very reliable and usually converges in at most two iterations. A randomized version of the method which reduce further the computational cost for large images problems is also proposed. Experimental results are provided and discussed.

Registration and Sequencing of Vessels Section Images at Macroscopic Levels

“…It has an opposite sign to (τ 1 , τ 2 ). Of course, there are known many other methods of translation estimation (see for example [14] and references cited therein). We have used the POC method, since it is relatively robust to the images differences.…”