In this paper, we consider multiobjective optimization for obtaining the feature weights associated with structural image impairments and its application to wireless imaging quality assessment. The proposed framework supports optimization with respect to finding a trade-off between metric prediction accuracy and generalization to unknown images. Evaluation of optimal trade-off solutions for two representative scenarios reveal the benefits of the proposed approach. In particular, quality prediction accuracy of an objective image metric can be strongly increased and negligible features are identified that can be discarded to save computational complexity.