Process monitoring using imaging can provide valuable information. However, the large number of images obtained necessitate automated classification into those showing “good” and “bad” product. This paper shows how a database of reference images can be used to modify image quality so as to obtain extremely high classification accuracies. The final model obtained combined adaptive image quality improvement with adaptive Bayesian Classification. Adapting to changes in image quality was accomplished with the aid of case-based reasoning. Experimental verification involved monitoring the presence or absence of purposefully added contaminant particles during extrusion of polyethylene. The new model was a major improvement over previous work and conclusively demonstrated the advantage of being able to adapt both image quality and the classification model itself to improve classification performance in dynamic environments involving large changes in image quality. Adaptation could be readily accomplished but did require human intervention to identify the need for adaptation and to accomplish it by using images of known class.