25th Signal Processing and Communications Applications Conference, SIU 2017, Antalya, Türkiye, 15 - 18 Mayıs 2017
Mutual Information (MI) is one of the main methods for efficient registration of multiband images in the literature. Since images on different bands are often expressed in different numbers of bits, contrast enhancement is inevitable before MI-based image registration. Although the contrast enhancement method used has a significant effect on the registration performance due to MI metric, this problem is not sufficiently addressed in the literature. In this paper, the effect of the outstanding contrast enhancement methods is examined on image registration performance. For this purpose, 16-bit thermal and 8-bit visible band satellite images were used and Monte Carlo tests were performed. First, random rotation and translation is applied to the image which is converted from 16-bit to 8-bit by contrast enhancement methods, then the transformation is tried to be estimated with MI and Multi-Level Single-Linkage (MLSL) optimization algorithm. Consequently, it is found that contrast enhancement methods have an effect on MI based image registration. Contrast limited adaptive histogram equalization (CLARE) and AHE methods have the highest registration performance according to the tests performed.