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 robot's goals. Two fundamental goals are determin-ing 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 index-ing 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 called Histogram Intersection, which matches model and im-age 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 called Histogram Backprojection, which performs this task efficiently in crowded scenes. 1

This article introduces a new indexing technique based on boundary histograms. For multicolored objects boundary histograms record estimates of the boundary lengths between different discrete colors in an image. Boundary histograms are small and insensitive to noise. To identify images we apply a match function to their boundary histograms. The match function that we derive handles occlusions and distracting pixels in the background of an object gracefully. The robustness and the low complexity of the match function together with its ability to distinguish many objects allow us to use boundary histograms as an index for large image databases. Test results illustrate the above mentioned features of boundary histograms and the match function. Keywords: Object recognition, color indexing, image signature, boundary length estimates. 1 Introduction Recent advances in data storage and data compression will allow image databases to explode in size. However, currently there exist only very ...

Since colour characterizes local surface properties and is largely viewpoint insensitive it is a useful cue for object recognition. Indeed, Swain and Ballard have developed a simple scheme, called colour-indexing, which identifies objects by matching colourspace histograms. Their approach is remarkably robust in that variations such as a shift in viewing position, a change in the scene background or even object deformation degrade recognition only slightly. Colour-indexing fails, however, if the intensity or spectral characteristics of the incident illuminant varies. This thesis examines two different strategies for rectifying this failure. Firstly we consider applying a colour constancy transform to each image prior to colour-indexing (colours are mapped to their appearance under canonical lighting conditions). To solve for the colour constancy transform assumptions must be made about the world. These assumptions dictate the types of objects which can be recognized by colour-indexing...

The goal of color constancy is to take the color responses (for example camera rgb triplets) of surfaces viewed under an unknown illuminant and map them to illuminant independent descriptors. In existing theories this mapping is either a general linear 3 x 3 matrix or a simple diagonal matrix of scaling coefficients. The general theories have the advantage that the illuminant can be accurately discounted but have the disadvantage that nine parameters must be recovered. Conversely while the coefficient theories have only three unknowns, a diagonal matrix may only partially discount the illuminant. My staring point in this thesis is to generalize the coefficient approach; the goal is to retain its inherent simplicity while at the same time increasing its expressive power. Under the generalized coefficient scheme, I propose that a visual system transforms responses to a new sensor basis before applying the scaling coefficients. I present methods for choosing the best coefficient bas...

This research is a study of the role of fixation and visual attention in object recognition. In this project, we built an active vision system which can recognize a target object in a cluttered scene efficiently and reliably. Our system integrates visual cues like color and stereo to perform figure/ground separation, yielding candidate regions on which to focus attention. Within each image region, we use stereo to extract features that lie within a narrow disparity range about the fixation position. These selected features are then used as input to an Alignment-style recognition system. We show that visual attention and fixation significantly reduce the complexity and the false identifications in model-based recognition using Alignment methods. We also demonstrate that stereo can be used effectively as a figure/ground separator without the need for accurate camera calibration. Thesis Supervisor: Eric Grimson Title: Associate Professor of Computer Science Acknowledgments I would lik...

The colors a camera sees depends on the color of the viewing illuminant and, unless corrected, this leads to poor color reproduction. There are two strategies to color correction: constrained and unconstrained. The constrained methods rely on simplifying assumptions about the world (e.g. there is a white reflectance is in every scene). If the assumptions hold then good correction is possible but if they do not hold, which is often the case, then poor correction results. In contrast, the unconstrained approach attempts to correct colors without making any world assumptions whatsoever. Rather, correction proceeds by exploiting only the information inherent in the physics of color image formation. Recent work by Finlayson has demonstrated that the unconstrained approach can deliver good color correction. Indeed excellent color correction has been demonstrated for images of everyday scenes. However, Finlayson's algorithm is relatively complex (high computational complexity) and it is also...

Abstract not found