Showing papers on "Scale-invariant feature transform published in 1999"

Object recognition from local scale-invariant features

Abstract: An object recognition system has been developed that uses a new class of local image features. The features are invariant to image scaling, translation, and rotation, and partially invariant to illumination changes and affine or 3D projection. These features share similar properties with neurons in inferior temporal cortex that are used for object recognition in primate vision. Features are efficiently detected through a staged filtering approach that identifies stable points in scale space. Image keys are created that allow for local geometric deformations by representing blurred image gradients in multiple orientation planes and at multiple scales. The keys are used as input to a nearest neighbor indexing method that identifies candidate object matches. Final verification of each match is achieved by finding a low residual least squares solution for the unknown model parameters. Experimental results show that robust object recognition can be achieved in cluttered partially occluded images with a computation time of under 2 seconds.

On the inverse Hough transform

Abstract: In this paper, an inverse Hough transform algorithm is proposed. This algorithm reconstructs correctly the original image, using only the data of the Hough transform space and it is applicable to any binary image. As a first application, the inverse Hough transform algorithm is used for straight-line detection and filtering. The lines are detected not just as continuous straight lines, which is the case of the standard Hough transform, but as they really appear in the original image, i.e., pixel by pixel. To avoid the quantization effects in the Hough transform space, inversion conditions are defined, which are associated only with the dimensions of the images. Experimental results indicate that the inverse Hough transform algorithm is robust and accurate.

Off-line signature recognition using parameterized Hough transform

Abstract: This article describes a method of an off-line signature recognition by using the Hough transform to detect stroke lines from the signature image. The Hough transform is used to extract the parameterized Hough space from the signature skeleton as a unique characteristic feature of signatures. In the experiment, the backpropagation neural network is used as a tool to evaluate the performance of the proposed method. The system has been tested with 70 test signatures from different persons. The experimental results reveal a recognition rate 95.24%.

Coarse-to-Fine Search Technique to Detect Circles in Images

Abstract: The aim of this paper is to propose an efficient coarse-to-fine search technique to reduce the storage and computing time in detecting circles in an image. Variable-sized images and accumulator arrays are used to reduce the computing and storage requirements of the Hough transform. The accuracy and the rate of convergence of the parameters at different iterations of the algorithm are presented. The results demon-strate that the coarse-to-fine search strategy is very suitable for detecting circles in real-time environments having time constraints.

Hough transform network

Abstract: A two-layered Hough transform network is proposed which accepts image co-ordinates as the input and learns the parametric forms of the lines in the image adaptively. It provides an efficient representation of visual information embedded in the connection weights. It not only reduces the large space requirements, as in the case of the classical Hough transform, but also represents the parameters with high precision.

An inherent probabilistic aspect of the Hough transform

Abstract: In this paper, a new property of the Hough transform is discovered, namely an inherent probabilistic aspect which is independent of the input image and embedded in the transformation process from the image space to the parameter space It is shown that such a probabilistic aspect has a wide range of implications concerning the specification of implementation schemes and the performance of Hough transform In particular, it is shown that in order to make the Hough transform really meaningful, an appropriate curve (surface) density function must be, either explicitly or implicitly, supplied during its implementation process, and that the widely used approach to uniformly discretizing parameter space in the literature is generally inadequate

Parameter space decomposition approach

Abstract: The Hough transform has been widely used in technique for geometric primitive extraction. However, one of the main problems of the Hough transform is its high cost of space, which greatly circumscribed its further applications. Not only does the standard Hough transform have such defect, but also the newly proposed Hough techniques such as randomized Hough transform, probabilistic Hough transform and dynamic Hough transform all suffer from the high space cost which has not been wonderfully solved. This paper proposes the parameter space decomposition approach to alleviate the high memory requirement. The basic principle of the parameter space decomposition approach is to use several 2 D arrays to implement a high dimension parameter space, which can drastically reduce the space burden. In fact the parameter decomposition approach can be considered as a trade off between a large space reduction and a slight false extraction rate. Numerous experiments show that the parameter space decomposition approach is an effective way of randomized Hough transform implementation.