What does your image look like?
You don't want a structuring element the size of your entire image! That will give you an image out that's just a single uniform gray level. Make your structuring element smaller, like 5 or 9 pixels in diameter.
What's all the stuff about binary image for?
Do you have a "blank shot"? If so, that would be better than using adapthisteq() which is really a last resort if you don't have a background image at all and need to estimate one from the image itself, assuming that some part of the background "shows through" your sample and you know that your sample is either brighter or darker than the background.
After you have the background image what you do depends on your image. For radiology and fluorescence you subtract the background, for most other image modalities you divide by the background.
