The paper presents a generic objectness measure that quantifies how likely an image window is to contain an object of any class. The measure is trained to distinguish objects with well-defined boundaries (e.g., cows, telephones) from amorphous background elements (e.g., grass, road). It combines several image cues, including a novel cue for closed boundaries, to assess objectness. The measure is evaluated on the challenging PASCAL VOC 07 dataset, where it outperforms state-of-the-art saliency measures and other methods. The objectness measure is applied to reduce the number of windows evaluated by class-specific object detectors and to reduce false positives. It is also shown to be useful in various applications, including weakly supervised learning, unsupervised segmentation, and object tracking. The measure is efficient, taking only about 4 seconds per image. The paper also introduces several image cues, including multi-scale saliency, color contrast, edge density, superpixel straddling, and location and size, which are learned and combined in a Bayesian framework to improve object detection. The results show that the combined objectness measure performs better than any individual cue and is effective in detecting objects in challenging scenarios.The paper presents a generic objectness measure that quantifies how likely an image window is to contain an object of any class. The measure is trained to distinguish objects with well-defined boundaries (e.g., cows, telephones) from amorphous background elements (e.g., grass, road). It combines several image cues, including a novel cue for closed boundaries, to assess objectness. The measure is evaluated on the challenging PASCAL VOC 07 dataset, where it outperforms state-of-the-art saliency measures and other methods. The objectness measure is applied to reduce the number of windows evaluated by class-specific object detectors and to reduce false positives. It is also shown to be useful in various applications, including weakly supervised learning, unsupervised segmentation, and object tracking. The measure is efficient, taking only about 4 seconds per image. The paper also introduces several image cues, including multi-scale saliency, color contrast, edge density, superpixel straddling, and location and size, which are learned and combined in a Bayesian framework to improve object detection. The results show that the combined objectness measure performs better than any individual cue and is effective in detecting objects in challenging scenarios.