If the vector layer contains the results of a previous classification (T1) and the image reflects new/current ground conditions (T2), then you could start by

1. create image objects (segmentation) - most likely multiresolution segmentation, but make sure to use the option to maintain vector outlines, then you will not only have image objects that reflect the input layers, but also the vector layer.
2. derive the object class from the vector layer - assign class by thematic attribute
3. do you just want to find change? - if so, you need to start defining that change given the input data you have (i.e. an object classified as bare soil in the vector that has a high NDVI in the image, or a water body that shows a low NDWI, etc.)
4. once the change objects are identified, you have the option of taking the analysis to the next step - what did they change to? Here you can limit the analysis on the class "change" and use descriptions for the target classes to assign them properly (i.e. that object that bare soil object - depending on the environment and temporal difference between T1 and T2 maybe you can you use some logic - could it go from bare soil to forest in 1 year?)

Hope this helps as guide you in the right direction!

Thanks for the description....