General object recognition and image understanding is recognized as a dramatic goal for computer vision and multimedia retrieval. In spite of the great efforts devoted in the last two decades, it still remains an open problem. In this paper, we propose a selective attention-driven model for general image understanding, named GORIUM (general object recognition and image understanding model). The key idea of our model is to discover recurring visual objects by selective attention modeling and pairwise local invariant features matching on a large image set in an unsupervised manner. Towards this end, it can be formulated as a four-layer bottomup model, i.e., salient region detection, object segmentation, automatic object discovering and visual dictionary construction. By exploiting multi-task learning methods to model visual saliency simultaneously with the bottom-up and top-down factors, the lowest layer can effectively detect salient objects in an image. The second layer exploits a simple yet effective learning approach to generate two complementary maps from several raw saliency maps, which then can be utilized to segment the salient objects precisely from a complex scene. For the third layer, we have also implemented an unsupervised approach to automatically discover general objects from large image set by pairwise matching with local invariant features. Afterwards, visual dictionary construction can be implemented by using many state-of-the-art algorithms and tools available nowadays.

Visual saliency plays an important role in various video applications such as video retargeting and intelligent video advertising. However, existing visual saliency estimation approaches often construct a unified model for all scenes, thus leading to poor performance for the scenes with diversified contents. To solve this problem, we propose a multi-task rank learning approach which can be used to infer multiple saliency models that apply to different scene clusters. In our approach, the problem of visual saliency estimation is formulated in a pair-wise rank learning framework, in which the visual features can be effectively integrated to distinguish salient targets from distractors. A multi-task learning algorithm is then presented to infer multiple visual saliency models simultaneously. By an appropriate sharing of information across models, the generalization ability of each model can be greatly improved. Extensive experiments on a public eye-fixation dataset show that our multi-task rank learning approach outperforms 12 state-of-the-art methods remarkably in visual saliency estimation.