We present a method to significantly reduce the complexity of polynomial-time object recognition algorithms while guaranteeing high tolerance to noisy and clutter data. Our approach uses a middle-level representation based on sparse groups, a novel concept introduced in this paper This representation permits to break down the recognition algorithm into two (or more) simpler stages, where each stage works with only a portion of the image features. This way of handling image information leads the algorithm to a considerably reduction of complexity. We implement a geometric matching based on boundary points as features, with a polynomial complexity.

We present a two-stage approach, Curved Inertia Frames, to compute texture groups. In the first stage, we use a filter-based approach to estimate central texture group locations. We use existing bank filters but suggest a new filter post-processing operation. In the second stage, we compute global texture groups using the estimates of the first stage. This is done using dynamic programming on a novel non-cartesian network. Our scheme is novel because it does not compute texture edges as an intermediate step, it uses a new post-processing operation, and because it is guaranteed to find a global optimum due to the non-cartesian network used in the second stage.

The authors present a novel vector ridge detector which automatically finds the right scale of a ridge even in the presence of noise, multiple steps and narrow valleys. One of the key features of the ridge detector is that it has a zero response at discontinuities. The ridge detector can be applied both to scalar and vector quantities such as color. A parallel perceptual organization scheme that works without edges and is based on the ridge detector is also presented. In addition to perceptual groups, the scheme computes potential focus of attention points at which to direct future processing. The relation to human perception and several theoretical findings supporting the scheme are presented. Results of a Connection Machine implementation of the scheme are presented for perceptual organization using color.>

The author introduces curved inertia frames (CIFs), a new definition of curved axes of inertia. A scheme is given for finding a frame of reference of a shape in an image based on such a definition and the author discusses how the frame can be used to describe the shape. The scheme assigns a saliency measure to each component of the reference frame that is a measure of its relevance, so that large and central parts play a more central role in the description of the shape. The scheme also computes a major axis that is used to organize the description of the shape, so that a canonical description can be obtained. One of the features of the scheme is its tolerance to noisy and spurious data. Several perceptual phenomena observed in humans such as grouping based on symmetry and environment bias in shape description can be reproduced naturally in this scheme.>