Transient Optics and

Real-world scene recognition has been one of the most challenging research topics in computer vision, due to the tremendous intra-class variability and the wide range of scene categories. In this paper, we successfully apply an evolutionary methodology to au-tomatically synthesize domain-adaptive holistic descriptors for the task of scene recognition, instead of using hand-tuned descriptors. We address this as an optimization problem by using multi-objective genetic programming (MOGP). Specifically, a set of primitive op-erators and filters are first randomly assembled in the MOGP frame-work as tree-based combinations, which are then evaluated by t-wo objective fitness criteria i.e., the classification error and the tree complexity. Finally, the best-so-far solution selected by MOGP is regarded as the (near-)optimal feature descriptor for scene recog-nition. We have evaluated our approach on three realistic scene datasets: MIT urban and nature, SUN and UIUC Sport. Exper-imental results consistently show that our MOGP-generated de-scriptors achieve significantly higher recognition accuracies com-pared with state-of-the-art hand-crafted and machine-learned fea-tures