Visual attention and target detection in cluttered natural scenes

Rather than attempting to fully interpret visual scenes in a fashion, biological systems appear to employ a serial strategy by an attentional spotlight rapidly selects circumscribed regions in the for further analysis. The spatiotemporal deployment of attention been shown to be controlled by both bottom-up (image-based) and -down (volitional) cues. We describe a detailed neuromimetic computer of a bottom-up scheme for the control of visual , focusing on the problem of combining information across modalities (orientation, intensity, and color information) in a purely stimulusdriven . We have applied this model to a wide range of target tasks, using synthetic and natural stimuli. Performance has, , remained difficult to objectively evaluate on natural scenes, no objective reference was available for comparison. We predicted search times for our model on the Search–2 database rural scenes containing a military vehicle. Overall, we found a poor between human and model search times. Further analysis, , revealed that in 75% of the images, the model appeared to the target faster than humans (for comparison, we calibrated the ’s arbitrary internal time frame such that 2 to 4 image locations visited per second). It seems that this model, which had originally designed not to find small, hidden military vehicles, but rather to the few most obviously conspicuous objects in an image, performed an efficient target detector on the Search–2 dataset. Further developments the model are finally explored, in particular through a more treatment of the difficult problem of extracting suitable low-level to be fed into the saliency map.