Showing papers on "Scale-invariant feature transform published in 1993"
28 Mar 1993
TL;DR: To detect shapes in noisy data, the fuzzy Hough transform is introduced, which finds shapes by approximately fitting the data points, which avoids the spurious shapes detected when using the conventional Houghtransform.
Abstract: To detect shapes in noisy data, the fuzzy Hough transform is introduced. This technique finds shapes by approximately fitting the data points, which avoids the spurious shapes detected when using the conventional Hough transform. An efficient implementation of this method is described for detecting lines and circles. >
59 citations
15 Jun 1993
TL;DR: Line features in geometric hashing are discussed, which is a technique for model-based object recognition in seriously degraded single intensity images and is shown that the technique is noise resistant and suitable in an environment containing many occlusions.
Abstract: Line features in geometric hashing are discussed. Lines are used as the primitive features to compute the geometric invariants, combining the Hough transform with a variation of geometric hashing as a technique for model-based object recognition in seriously degraded single intensity images. The effect of uncertainty of line features on the computed invariants for the case where images are formed under affine viewing transformations is analytically determined. The system is implemented with experiments on polygonal objects, which are modeled by lines. It is shown that the technique is noise resistant and suitable in an environment containing many occlusions. >
11 citations
13 Sep 1993
TL;DR: The basic, earlier version of the method, called the Motion Detection using Randomized Hough Transform (MDRHT), utilizes edge points as its features, but this version is extended to use both edge pixels and more local information of the edge pixels.
Abstract: Developments of the Hough Transform have led to new possibilities of motion detection. A new and robust Hough Transform, called the Randomized Hough Transform (RHT), has been applied to motion analysis. The basic, earlier version of the method, called the Motion Detection using Randomized Hough Transform (MDRHT), utilizes edge points as its features. In this paper, the MDRHT is extended to use both edge pixels and more local information of the edge pixels, e.g., gradients of the edge pixels. Two novel methods are proposed, tested, and compared with the earlier version of the MDRHT.
7 citations
13 Sep 1993
TL;DR: This paper proposes the novel use of Fourier descriptor as parameterized curve equation in a Hough transform and shows that using this Fourier parameterization, all parameters are bounded and the representation have uniform accuracy for all parameters.
Abstract: The Hough transform is a popular technique for feature detection and object recognition. In its original formulation for lines, the slope parameter is unbounded. Moreover, a rotation of the image will cause uneven representation accuracy of the slope. In a classic paper, Duda and Hart solved these problems by introducing a modified ρ-Θ parameterization, which has since become very popular in the computer vision community for parameterizing lines. Unfortunately, no equivalent parameterization exists for curves. This has led to ad hoc choice of parameterizations for circles, ellipses, conic sections etc. In this paper, we propose the novel use of Fourier descriptor as parameterized curve equation in a Hough transform. We show that using this Fourier parameterization, all parameters are bounded and the representation have uniform accuracy for all parameters.
6 citations
07 May 1993
TL;DR: This article considers how to keep the robustness of the Hough line finding algorithm, without sacrificing the accuracy in line parameters which may be needed for higher levels of processing, and without making the algorithm so slow as to be impracticable.
Abstract: This article considers how to keep the robustness of the Hough line finding algorithm, without sacrificing the accuracy in line parameters which may be needed for higher levels of processing, and without making the algorithm so slow as to be impracticable. >
4 citations
14 Jul 1993
TL;DR: The role of the image processing techniques in this algorithm, specifically, the application of the Hough Transform in this scheme is described.
Abstract: This paper describes a vision based vehicle identification algorithm. This algorithm has been developed using image processing and pattern recognition techniques. The paper describes the role of the image processing techniques in this algorithm, specifically, the application of the Hough Transform in this scheme.
4 citations
27 Apr 1993
TL;DR: By making use of the Hough and the contour sequence matching technique, the authors suggest a novel approach for solving the problem of the large memory requirement of theHough space.
Abstract: By making use of the Hough and the contour sequence matching technique, the authors suggest a novel approach for solving the problem of the large memory requirement of the Hough space. For the conventional Hough algorithm it is necessary to take a four-dimensional space for the recognition of an irregular pattern, while in the proposed algorithm the Hough space is replaced with a two-dimensional domain. The orientation of a possible object is determined by comparing the contour sequence of the object and that of the prototype. The reduction of the scaling and orientation parameters in the new Hough space also simplifies the peak searching process to verify the results. In order to enhance the accuracy of the recognition rate, a novel peak searching technique for eliminating false peaks in the Hough space is suggested. The overall approach was tested using practical images and was found to be very promising. >
3 citations
13 Sep 1993
TL;DR: A method to extract 3D information by projecting back the feature points in the images into the 3D space by using the Hough transform technique, called the double backprojection method (DBP).
Abstract: We propose a method to extract 3D information by projecting back the feature points in the images into the 3D space. We divide the 3D space into voxels and apply the Hough transform technique by giving a vote to every voxel on the backprojected lines. However, a simple voting rule raises some problems. We propose a wellcontrived voting and evaluation rule to solve these problems, which we call the double backprojection method (DBP). The octree representation of objects can be adopted to DBP to allow multi-resolution analysis and to increase calculation efficiency. Experimental results are also described.
1 citations
07 May 1993
TL;DR: In this paper, the location of arbitrary nonanalytic shapes at random orientations using only one plane of parameter space is described, where the orientation of the sought-after shape is unknown.
Abstract: The storage requirement, and hence the computational load, of the generalised Hough transform increases considerably when the orientation of the sought-after shape is unknown. A method is described which allows the location of arbitrary nonanalytic shapes at random orientations using only one plane of parameter space.<
>
1 citations
26 Jul 1993
TL;DR: Two kinds of Hough transform are optically implemented by a two-dimensional array of computer-generated-holograms which simultaneously reconstruct two point spread functions corresponding to lines and circles.
Abstract: Optical implementation of Hough transform filter is investigated for parallel detecting two kinds of element shape. Two kinds of Hough transform are optically implemented by a two-dimensional array of computer-generated-holograms which simultaneously reconstruct two point spread functions corresponding to lines and circles. A novel matrix format is adopted to parameter domain in Hough transform instead of conventional orthogonal coordinate system.
09 Apr 1993
TL;DR: The efficient utilization of the Hough Transform to further classify the properties components of an image is presented here for an n X n image.
Abstract: The detection of lines and curves in binary and gray level images is based in the Hough Transform. This detection is a vital first step in the detection and classification of shapes in images. The Linear Array with Reconfigurable Global Bus System has been shown to efficiently accomplish a number of low and medium level image processing tasks including the identification of connected components in an image. The efficient utilization of the Hough Transform to further classify the properties components of an image is presented here for an n X n image. The Hough transform is accomplished in O(k (theta) log n) time using n 2 processing elements. Where k (theta) is the number of angles in the (theta) search space. We also examine a customization of the Hough Transform specifically adapted for the Linear Array with Reconfigurable Global Bus System.