Showing papers on "Scale-invariant feature transform published in 1990"

Model-based object recognition by geometric hashing

Abstract: The Geometric Hashing paradigm for model-based recognition of objects in cluttered scenes is discussed. This paradigm enables a unified approach to rigid object recognition under different viewing transformation assumptions both for 2-D and 3-D objects obtained by different sensors, e.g. vision, range, tactile. It is based on an intensive off-line model preprocessing (learning) stage, where model information is indexed into a hash-table using minimal, transformation invariant features. This enables the on-line recognition algorithm to be particularly efficient. The algorithm is straightforwardly parallelizable. Initial experimentation of the technique has led to successful recognition of both 2-D and 3-D objects in cluttered scenes from an arbitrary viewpoint. We, also, compare the Geometric Hashing with the Hough Transform and the alignment techniques. Extensions of the basic paradigm which reduce its worst case recognition complexity are discussed.

The Dynamic Generalized Hough transform

Abstract: A new algorithm for computing the Hough transform has been presented. It uses information present in the location of the feature points to reduce the generation of evidence in the transform plane. The algorithm gives improved performance compared with the standard Hough transform. The improvement is in computation time and memory allocation. Further advantages of using the algorithm are that peak detection is one dimensional and the end points of curves may be detected. The algorithm is also inherently parallel.

A probabilistic approach to the Hough Transform.

Abstract: The mathematical principles come from the "Maximum Likelihood Method" [2, 3], used in probability theory for the determination of distribution parameters from experimental data. The Maximum Likelihood analysis leads to the definition of the Probabilistic Hough Transform, which is a likelihood function. If certain assumptions are made about the error characteristics, the PHT is very close to conventional Hough Transforms. If, in a particular application, these assumptions are a reasonable approximation, good results are usually obtained using standard Hough methods. However, where these assumptions are far from the truth, the Hough Transform will not work well, and steps should be taken to improve the model of input feature errors, such as filtering the Hough space, or incrementing an extended region instead of just the voting space. As a last resort, the full PHT can be computed, but this is much more computationally expensive than conventional Hough methods.

Performance improvements to the Hough transform

Abstract: The paper describes a new algorithm for the Hough transform, the combinatorial Hough transform, an improvement to the Hough space accumulator which gives greater resolution for a fixed size, and a new method for detecting line end points using the transform. These are described separately and implemented together to give a new algorithm with improved resolution. Also a method for the performance analysis of Hough transforms is introduced and exercised on the new algorithms, comparing their error performance with that of the standard Hough transform algorithm. It is found that the new algorithm with the parameter space improvements offers some real advantages over the standard and combinatorial algorithms, though this is dependent on the precise details of the implementation. The error analysis procedure is described in some detail and its effectiveness is discussed together with desirable enhancements.

FIHT2 Algorithm: A Fast Incremental Hough Transform.

Abstract: FIHT2 algorithm defined by p = x . cos 0 + y .sin 0 + (a/(21()) . x. sine at 0 5 6 < a/2 and at p = x. cose + y . sin 0 + (aJ(2It')) . y . cos 0 at a/2 5 0 < a is a Hough transform which requires nothing of the trigonometric and functional operations to generate the Hough distributions. It is demonstrated in this paper that the FIHT2 is a complete alternative of the usual Hough transform(HT) defined by p = x.cos O+y.sin 0 in the sense that the both transforms could work perfectly as a line detector. It is easy to show that the Hough curves of the FIIJT2 can be generated in a incremental way where addition operation is exclusively needed. It is also investigated that the difference between HT and FIHT2 could be estimated to be neglected.

Dynamic generalized Hough transform

Abstract: A new algorithm for the Generalized Hough transform is presented The information available in the distribution of image points is used to optimize the computation of the transform The calculated parameters are those associated with a single image point and all other image points in combinations of the minimum number of points necessary to define an instance of the shape under detection The method requires only one dimensional accumulation of evidence Using the algorithm, the transform of sparse images is more efficiently calculated Dense images may be segmented and similarly processed In two dimensions, the method provides a feedback mechanism between image and transform space whereby contiguity of feature points and endpoints of curves may be determined© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering Downloading of the abstract is permitted for personal use only

The modified adaptive Hough transform (MAHT)

Abstract: A fast and robust algorithm for extracting line segments from an image through use of the Hough Transform is presented. The framework for our algorithm comes from the work of Illingworth and Kittler. We have extended their Adaptive Hough Transform algorithm to handle lines of arbitrary slope, multiple lines, and lines in very low signal to noise environments. We present our Modified Adaptive Hough Transform along with experimental results obtained with data from a tactile sensor.

On extending the generalized hough transform

Abstract: Matching models to images is a very important task in image understanding. The generalized Hough transform (GHT) proposed by Ballard has been proven to be a very effective method for matching models of arbitrary shapes to images. It converts the problem of global pattern detection into a problem of local peak finding. However, the GHT has difficulty handling images with occlusions due to the reduced peak values when objects are partially occluded. It can easily make mistakes when images contain patterns similar to the model. Moreover, it is also very difficult to apply parallel processing. Thc unique peak spot results in contention when processors try to access it. In this thesis, An improved version of Ballard's GHT is proposed which provides a potentially more robust and systematic technique, the linear Hough transform, for solving the problems in the detection of partially occluded objects. We use a linear numeric pattern to replace the peak in the GHT and use the relationship between entries in the linear pattern to achieve high robustness. Partial matches of the linear pattern are used for partial object detection. Finally, we present a parallel version of our new technique to exploit the parallel computational power of array processors. High performance is achieved by taking advantage of the speed-up due to reduced contention in the accumulation process which results with our linear numeric pattern.