Five important trends have emerged from recent work on computational models of focal visual attention that emphasize the bottom-up, image-based control of attentional deployment. First, the perceptual saliency of stimuli critically depends on the surrounding context. Second, a unique 'saliency map' that topographically encodes for stimulus conspicuity over the visual scene has proved to be an efficient and plausible bottom-up control strategy. Third, inhibition of return, the process by which the currently attended location is prevented from being attended again, is a crucial element of attentional deployment. Fourth, attention and eye movements tightly interplay, posing computational challenges with respect to the coordinate system used to control attention. And last, scene understanding and object recognition strongly constrain the selection of attended locations. Insights from these five key areas provide a framework for a computational and neurobiological understanding of visual attention.

/pdf/computational-modelling-of-visual-attention-3wtbok1ou7.pdf

Computational modelling of visual attention.

http://ilab.usc.edu/publications/doc/Itti_Koch00vr.pdf

A saliency-based search mechanism for overt and covert shifts of visual attention.

IEEE Transactions on Pattern Analysis and Machine Intelligence

The gaze movements accompanying target localization were examined via human observers and a computational model (target acquisition model [TAM]). Search contexts ranged from fully realistic scenes to toys in a crib to Os and Qs, and manipulations included set size, target eccentricity, and target-distractor similarity. Observers and the model always previewed the same targets and searched identical displays. Behavioral and simulated eye movements were analyzed for acquisition accuracy, efficiency, and target guidance. TAM's behavior generally fell within the behavioral mean's 95% confidence interval for all measures in each experiment/condition. This agreement suggests that a fixed-parameter model using spatiochromatic filters and a simulated retina, when driven by the correct visual routines, can be a good general-purpose predictor of human target acquisition behavior.

A theory of eye movements during target acquisition.

A method that treats linear neighborhood operators within a unified framework that enables linear combinations, concatenations, resolution changes, or rotations of operators to be treated in a canonical manner is presented. Various families of operators with special kinds of symmetries (such as translation, rotation, magnification) are explicitly constructed in 1-D, 2-D, and 3-D. A concept of 'order' is defined, and finite orthonormal bases of functions closely connected with the operators of various orders are constructed. Linear transformations between the various representations are considered. The method is based on two fundamental assumptions: a decrease of resolution should not introduce spurious detail, and the local operators should be self-similar under changes of resolution. These assumptions merely sum up the even more general need for homogeneity isotropy, scale invariance, and separability of independent dimensions of front-end processing in the absence of a priori information. >

Generic neighborhood operators

The TNO Human Factors Searchu2 image dataset consists of: a set of 44 high-resolution digital color images of different complex natural scenes, the ground truth corresponding to each of these scenes, and the results of psychophysical experiments on each of these images. The images in the Searchu2 dataset are a subset of a larger set that has been used in a visual search and detection experiment. Each scene (image) contains a single military vehicle that serves as a search target. The image dataset, an Excel file with the ground truth and observer data, and an extensive report describing the dataset are available on CD-ROM (requests by email to the first author). The dataset can be used to develop and validate digital metrics that compute the visual distinctness (contrast, conspicuity, or saliency) of targets in complex scenes, and models of human visual search and detection. The dataset has already been used in more than ten different studies in the literature, ranging from studies evaluating target detectability metrics to eye movement studies and attempts to model the human visual system. In addition to this work, eight other articles in this special section address the Searchu2 dataset.

/pdf/image-dataset-for-testing-search-and-detection-models-1qljm4idjd.pdf

Image dataset for testing search and detection models

Een nieuwe methode voor het karakteriseren van prestaties met electro- optische waarnemingsapparatuur wordt voorgesteld. De methodiek heeft een groot aantal voordelen boven de huidige NATO standaarden, de MRTD en de MRC, en voorspelt bovendien voor vele doelen veel beter.

Triangle orientation discrimination: the alternative to minimum resolvable temperature difference and minimum resolvable contrast

The Triangle Orientation Discrimination (TOD) threshold is an alternative method to characterize electro-optical system performance. The method is based on an improved test pattern, a better-defined observer task and a solid psychophysical measurement procedure, and allows accurate and comparable results from different measuring teams. With the TOD methodology, accurate sensor performance measuring and Go-NoGo testing can be easily performed.

Guidelines for accurate TOD measurement

A simple and efficient psychophysical procedure is presented to quantify the visual conspicuity of a target in a complex (natural) scene. Measurements can easily and quickly be performed in the field or in complex environments. Only a few observers (typically two to three) are needed to achieve sufficient accuracy. We show that this conspicuity measure predicts human visual search performance in realistic and military relevant complex scenarios. Also, conspicuity measured on photographic slides agrees with conspicuity measured in the field, which implies that the new conspicuity measure can be used in combination with photosimulation studies to optimize and evaluate the visual distinctness of displayed information.

Visual conspicuity determines human target acquisition performance

The standard way to characterize sensor performance is by means of the Minimum Resolvable Temperature Difference (MRTD) and Minimum Resolvable Contrast (MRC) methods. These methods are based on Fourier analysis and work reasonably well for linear (analogue) systems. However, nonlinear effects, such as sampling, are not properly accounted for. As an alternative, the Triangle Orientation Discrimination (TOD) method has been proposed, based on 4 oriented triangles, that can handle nonlinear effects. Here, we present a model that predicts the TOD-sensor performance characterization curve from the system parameters. It consists of i) a sensor model and ii) a model of the visual system of the observer. The sensor model generates display images which are fed into the visual system model. The visual system is modeled by a bank of band-pass filters which mimic the pattern of neural activity in the visual cortex (using a common stack model). The neural activity is calculated and internal noise is added. Finally, a decision is made based on a correlation with the expected neural activity of the 4 possible inputs. The model has been validated with two human observer experiments in which the TOD-curve 1) of the naked eye, and 2) of a simulated thermal staring sensor system were measured. An internal noise level could be found for which the TOD of the naked eye of the two observers could be predicted. The model gives reasonable (but somewhat optimistic) predictions of the TOD-sensor performance curve of the simulated staring camera. Although more tests and modifications are required, these preliminary results suggest that the model can be developed into a model which predicts the TOD for all kinds of sensor systems, which may include sampling effects, noise, blur and (local) image enhancement methods.

/pdf/capturing-the-sampling-effects-a-tod-sensor-performance-zyx3x0e8s0.pdf

Piet Bijl

Papers

Image dataset for testing search and detection models

Triangle orientation discrimination: the alternative to minimum resolvable temperature difference and minimum resolvable contrast

Guidelines for accurate TOD measurement

Visual conspicuity determines human target acquisition performance

Capturing the sampling effects : A TOD sensor performance model