Michael J. Swain

Color indexing

Computer vision is embracing a new research focus in which the aim is to develop visual skills for robots that allow them to interact with a dynamic, realistic environment. To achieve this aim, new kinds of vision algorithms need to be developed which run in real time and subserve the robots goals. Two fundamental goals are determining the location of a known object. Color can be successfully used for both tasks.This article demonstrates that color histograms of multicolored objects provide a robust, efficient cue for indexing into a large database of models. It shows that color histograms are stable object representations in the presence of occlusion and over change in view, and that they can differentiate among a large number of objects. For solving the identification problem, it introduces a technique calledHistogram Intersection, which matches model and image histograms and a fast incremental version of Histogram Intersection, which allows real-time indexing into a large database of stored models. For solving the location problem it introduces an algorithm calledHistogram Backprojection, which performs this task efficiently in crowded scenes.

Indexing via color histograms

This paper shows color histograms to be stable object representations over change in view, and demonstrates that they can differentiate among a large number of objects. The authors introduce a technique called histogram intersection for efficiently matching model and image histograms. Color can also be used to search for the location of an object. An algorithm called histogram backprojection performs this task efficiently in crowded scenes.<<ETX>>

Promising directions in active vision

Active vision systems have mechanisms that can actively control camera parameters such as position, orientation, focus, zoom, aperture, and vergence (in a two-camera system) in response to the requirements of the task and external stimuli. They may also have features such as spatially variant (foveal) sensors. More broadly, active vision encompassesattention, selective sensing in space, resolution, and time, whether it is achieved by modifying physical camera parameters or the way data is processed after leaving the camera.In the active-vision paradigm, the basic components of the visual system are visual behaviors tightly integrated with the actions they support; these behaviors may not require elaborate categorical representations of the 3-D world. Because the cost of generating and updating a complete, detailed model of most environments is too high, this approach to vision is vital for achieving robust, real-time perception in the real world. In addition, active control of imaging parameters has been shown to simplify scene interpretation by eliminating the ambiguity present in single images. Important research areas in active vision include attention, foveal sensing, gaze control, eye-hand coordination, and integration with robot architectures.

Task and Environment-Sensitive Tracking,

Abstract : In a mobile robot, visual tracking, like other visual behaviors, takes place in a context that includes aspects of the task, the object being tracked, and the background. In this work, prior knowledge of those task and target characteristics that either enable or hinder different real-time image-tracking algorithms, together with run-time evaluation of the robots environment, are used to select an algorithm appropriate to the context.

Shape from patterns: Regularization

We present a theory for the recovery of the shape of a surface covered with small elements (texels). The theory is based on the apparent surface-pattern distortion in the image and fits the regularization paradigm, recently introduced in computer vision by Poggio et al., [1985]. A mapping is defined on the basis of measurement of the local distortions of a repeated unknown texture pattern due to the image projection. From this, a constraint on the gradient orientation is determined from the apparent area of a texture element. The analysis is done under an approximation of the perspective projection called paraperspective. The resulting algorithm is applied to several synthetic and real images to demonstrate its performance.

Texture, contour, shape, and motion

Abstract Intrinsic magic calculation exploits constraints arising from physical and imaging processes to derive physical scene parameters from input images. After a brief review of a paradigmatic intrinsic image calculation we turn to a recent result in shape from texture and then to a new result that derives shape and motion from a sequence of patterned inputs. Experimental results are demonstrated for synthetic and natural images.

Comments on Samal and Henderson: :20parallel consistent labeling algorithms”

Samal and Henderson claim that any parallel algorithm for enforcing arc consistency in the worst case must have Ω(na) sequential steps, wheren is the number of nodes, anda is the number of labels per node. We argue that Samal and Hendersons argument makes assumptions about how processors are used and give a counterexample that enforces arc consistency in a constant number of steps usingO(n[su2a22na) processors. It is possible that the lower bound holds for a polynomial number of processors; if such a lower bound were to be proven it would answer an important open question in theoretical computer science concerning the relation between the complexity classesP andNC. The strongest existing lower bound for the arc consistency problem states that it cannot be solved in polynomial log time unlessP=NC.

Efficient Parallel Estimation for Markov Random Fields

Abstract We present a new, deterministic, distributed MAP estimation algorithm for Markov Random Fields called Local Highest Confidence First (Local HCF). The algorithm has been applied to segmentation problems in computer vision and its performance compared with stochastic algorithms. The experiments show that Local HCF finds better estimates than stochastic algorithms with much less computation.

Object Identification and Search: Animate Vision Alternatives to Image Interpretation

We are accustomed to thinking of the task of vision as being the construction of a detailed representation of the physical world. However, a paradigm that we term animate vision argues that vision is more readily understood in the context of the tasks that the system is engaged in, and that these tasks may not require elaborate categorical representations of the 3-D world. As an example, we show how the general problem of image interpretation can be replaced in many cases by a combination of two simpler problems, identification and search. Both tasks use multidimensional color histograms to represent the model and images. Color histograms are shown to permit efficent matching and a sufficiently rich representation to distinguish among a large number of objects.