Simple explanation, both are methods to extract information from an image. Classification is the historical word for this (originally several ISODATA techniques for unsupervised and several supervised methods). These methods looked at digital number (spectral) similarities among pixels in the image. They work great on spatial resolutions larger than 10 meters. When you go to high resolution (extracting homes), the old methods lump all houses with the same thematic class. A post processing clump would break each house into a separate group. Sounds good, but there were a lot of mixed pixels on the homes, so the each house did not pop out of the image, nor separate very well.
Segmentation is a classification technique that also looks for similar characteristics of pixels, but uses more than only spectral characteristics. ERDAS IMAGINE 2013 uses spectral, connectivity (what is my neighbor pixel?), texture, size, shape and other characteristics to extract features from an image. This allows you to pull out a specific house.
Edited by pbeaty on 11/29/12 05:06 AM.