This paper studies the problem of combining region and boundary cues for natural image segmentation. We employ a large database of manually segmented images in order to learn an optimal affinity function between pairs of pixels. These pairwise affinities can then be used to cluster the pixels into visually coherent groups. Region cues are computed as the similarity in brightness, color, and texture between image patches. Boundary cues are incorporated by looking for the presence of an “intervening contour”, a large gradient along a straight line connecting two pixels. We first use the dataset of human segmentations to individually optimize parameters of the patch and gradient features for brightness, color, and texture cues. We then quantitatively measure the power of different feature combinations by computing the precision and recall of classifiers trained using those features. The mutual information between the output of the classifiers and the same-segment indicator function provides an alternative evaluation technique that yields identical conclusions. As expected, the best classifier makes use of brightness, color, and texture features, in both patch and gradient forms. We find that for brightness, the gradient cue outperforms the patch similarity. In contrast, using color patch similarity yields better results than using color gradients. Texture is the most powerful of the three channels, with both patches and gradients carrying significant independent information. Interestingly, the proximity of the two pixels does not add any information beyond that provided by the similarity cues. We also find that the convexity assumptions made by the intervening contour approach are supported by the ecological statistics of the dataset. 1.

We test hypotheses concerning human cue combination in a slant estimation task. Observers

this article online for further analysis.) study were more likely to use commonplace rather than Figure 4C shows the predicted and observed JNDs ad hoc strategies. The fact that nearly optimal cue inte- for small or zero conflicts for each observer and each gration was observed in all three studies suggests that stimulus orientation. The good agreement between prethe phenomenon is pervasive. dicted and observed shows that individual differences The observed and predicted PSEs in our experiment in intermodal discrimination can be largely explained by behavior in the within-modality experiments. were very similar (Figure 3D), but the observed and pre- 487 were otherwise transparent. Because element size and density were dicted JNDs differed consistently (Figures 4A and 4B). randomized, they were not a reliable cue to intersurface distance

Abstract not found

iewing multistable figures (reviewed in Leopold & Logothetis, 1999 and Blake & Logothetis, 2002) we understand some of the mechanisms responsible for perceptual selection. Several interpretations of a multistable stimulus are represented in the visual cortical activity concurrently, even though only one of the alternatives is perceived at a time. How can perceptual experience be stable and continuous in the presence of other interpretations? To answer this question, we must understand the interplay of two counteracting temporal tendencies in the perception of multistable figures: hysteresis and adaptation. Hysteresis increases the likelihood of the current percept in the next We are grateful to H. Hock and D. R. Proffitt for valuable discussions, and to W. Epstein, H. Hock and J. Wagemans for helpful suggestions about an early version of the manuscript. This work was supported by NEI Grant R01 EY 12926. Send correspondence to either author at 360 Minor Hall, Vision Science, University

Abstract not found

We propose a novel model-based coding system for video. Model-based coding aims at improving compression gain by replacing the non-informative image elements with some perceptually equivalent models. Images enclosing large textured regions are ideal candidates. Texture movies are obtained by filming a static texture with a moving camera. The integration of the motion information within the generative texture process allows to replace the "real" texture with a "visually equivalent" synthetic one, while preserving the correct motion perception. Global motion estimation is used to determine the movement of the camera and to identify the overlapping region between two successive frames. Such an information is then exploited for the generation of the texture movies. The proposed method for synthesizing 2D+1 texture movies is able to emulate any piece-wise linear trajectory. Compression performances are very encouraging. On this kind of video sequences, the proposed method improves the compression rate of an MPEG4 state-of-the-art video coder of an order of magnitude while providing a sensibly better perceptual quality. Importantly, the current implementation is real-time on Intel PIII processors.

family of distributions

similar in the parallel and perpendicular cases; so, in this situation, the precision of haptic estimates should not vary with orientation (see [3] for a counter example). Suppose the observer looks at and feels the surfaces simultaneously. The principle of maximum likelihood (ML) prescribes the strategy for combining visual and haptic estimates that produces the estimate of lowest variance [4–8]. If the visual and haptic estimates are independent and normally distributed, that strategy is weighted summation Vision and haptics have different limitations and advantages because they obtain information by different methods. If the brain combined information from the Sˆ VH � wVS ˆ V � wHS ˆ H, two senses optimally, it would rely more on the one

In estimating properties of the world, we often use multiple sources of information. For example, in estimating the 3-d layout of a scene, there are many sources of information or “cues ” available for the estimation of depth and shape (Kaufman, 1974). These include binocular cues (disparity, vergence), motion