Comparison of distance measures according to suitability for 2D electrophoresis image registration using synthetic image data and SOFM

摘要

This paper presents a comparative investigation of distance measures that may be used in image registration algorithms specialized for two-dimensional electrophoresis gel images. Standard image registration techniques employ correlation-based similarity measures but they are not robust against geometric and large intensity distortions. Exploitation of specific electrophoresis gel image patterns allows achieving better registration performance. In order to make comparison of distance measures synthetic electrophoresis images were generated. Synthetic data gave ability to control variation of amount of image dissimilarity specific to 2D electrophoresis images. Relation of distance measure response to variation of image dissimilarity was analyzed using self-organizing feature maps. Measures that gave desired responses of self-organizing feature maps were determined. These results were consistent with matching performance of real gel image regions. Computational needs of measures were tested additionally.