I use QGIS as my GIS program and this has a plug-in (Semiquatitative Classification Plug-in) that calculates the DN and reflectance values for Landsat data. I then use the corrected reflectance data in my band ratio calculations.
You can use DN if you are just interested in comparing band ratio values within a single image. If you want to compare band ratios from different images, especially if those images are collected on different date or with different sensors, it is best to use reflectance values to compute the band ratios. I'm attaching a link to a paper that discusses the methodology.
Yes you are right !!! When comparing two difffrent images it is better to process a rigourous atmospheric correction and it is not my case. Reflectance can be obtained by a simple calibration of the original image. Thank you for the paper.
Like Brian said, if you just have one sensor image you can use DN whitout any problem, and for other images processing before you convert to reflectance you must do the atmospheric correction for all the sensors.
If you have your image in reflectance, like you said, you can do it too.