<p>In recent years, the use of X-ray fluorescence (XRF) imaging has grown in the field of cultural heritage. This type of imaging can be complimentary to multi/hyperspectral imaging systems, providing the conservators with information that is useful to identify materials on paintings or manuscripts. Whereas hyperspectral imaging can provide information at the material molecular level, XRF provides information on the elements present. To extract the full information from the XRF element map (which is obtained after fitting the spectra), one should first register them to a reference image, generally a traditional high-resolution color image. This helps locate where exactly each element is present on the artifact, which can especially help the conservators with their conservation and even preservation efforts. Most methods that are used to overcome this issue rely on user input, in which a few reference points on both the XRF and RGB are selected and are then used for the registration process. This way the accuracy of the registration is dependent on the accuracy by which the user selects the points. In this work, a fully automatic method is developed for XRF image registration. First, the reference RGB image is converted into a grayscale image, helping to increase the similarity between these two different modalities. Then the XRF image that has the highest Structural Similarity Index (SSI) with respect to the reference image is specified. That image is then registered to the reference image using Maximization of Mutual Information (MMI). The same transformation is also used for the other XRF element maps in the dataset leading to all the images being registered. In cases, where none of the XRF element maps could be registered to the reference image due noise and/or low resolution, it is shown that quantization of the XRF element maps and the grayscale RGB reference image significantly improves performance. This process suppresses the noise and enhances some features resulting in accurate registration between the two differing modalities. It should be noted that this is the first time, to our best knowledge, that the quantization is applied to XRF images for the purpose of noise reduction and image registration. The proposed algorithm was applied to three different sets of XRF and RGB images in terms of resolution and noise level. These different sets showed the usefulness of the algorithm developed herein, in that the sets with good and decent quality XRF element maps did not need to be quantized as the MMI was able to register the original XRF to the original RGB. However, the set that is comprised of XRF element maps of very low resolution needed to undergo the process of quantization before being registered to the RGB image using MMI. The method proposed herein is faster and need little to no interaction with the user as opposed to the prior approach. It will also be shown in order to select a candidate XRF image to be registered with the RGB image, the initial value of SSI seems more efficient than the use of MI as used by the prior method in the literature.</p>