In the end, not every image can be segmented using thresholding. The best images for thresholding are those where there is a good separation between the objects of interest. There are many images which cannot be properly segmented by any means (the presence of colour in images does help, but is not a panacea for segmentation). Good examples of things that lead to reasonable thresholding results include such things as text images, B&W drawings, and images where objects can be easily differentiated. Below are some examples.
Choosing a global or localized thresholding algorithm really depends in the content you wish to segment, and whether or not it is compromised by other things in the image, e.g. non-uniform background, overlapping objects etc. There is no perfect thresholding algorithm, and some images are not optimal for turning into a binary image, or even an image with four sections. It would be like trying to segment the image below.