Do you really need/want to? Depending on your purpose:
a) if you only need approximate concentration, for example for cloning, easiest is to load a DNA ladder with a known amount of DNA per band - see here:
Load some dilutions of your sample and compare band intensities with the ladder.
b) if you need very accurate measurement of bands, agarose gels & ethidium bromide and then measuring band intensities are not a good choice because the commonly used camera imaging system which takes pictures and prints them will not accurately reflect band intensities when bands are overexposed (white will be white no matter how many more photons are emitted in the same space). Film simply hasn't got a wide dynamic range. This is also true for determining protein conc from exposed film Western Blots!!!
c) Anyway, unless you have a fancy new digital system that actually measures photon emission and gives you measurements directly, using band intensities means you underestimate your concentration. But you wouldn't need ImageJ then because those machines come with their own software...
d) If you can get access to a Bioanalyzer, you can use the DNA kit to determine conc. with 5-15% CVs depending on kit and size of fragments.
e) If you are planning to use analysis of band intensities to determine PCR product amount to determine RNA expression levels, this absolutely is not a good idea and will not get you good enough data!
If you are absolutely set on using ImageJ to quantify bands as best as you can, you need to do the following:
Load sample in different dilutions, load DNA ladder or control samples of known amount in dilutions too. Best if you can actually add a DNA fragment of known size and concentration to each sample as loading control as well.
All measured samples must be on the same gel to be comparable! Draw a square around your band of interest in samples&controls (should be same size for all of them) and plot the resulting measurements vs concentration (dilution factors for sample) for each sample and control on a graph. You should have linear plots for each and they should be parallel if you used the same dilution factors for sample&control. Then you can use the time honoured method of using your control concentration to determine your DNA conc. of interest.
All in all lots of work for data that's not great... so please do consider a) to e)