Mark Hedley

Fast corner detection

Abstract This paper describes a new corner detection algorithm, based on the property of corners that the change of image intensity should be high in all directions. Consequently, the corner response function (CRF) is computed as a minimum change of intensity over all possible directions. To compute the intensity change in an arbitrary direction an interpixel approximation is used. A multigrid approach is employed to reduce the computational complexity and to improve the quality of the detected corners. This algorithm, and other popular corner detectors, were evaluated and compared on the basis of their consistency, accuracy and speed using a range of images and video sequences. It was found that our algorithm performs well compared to the other algorithms, but it is significantly faster to compute.

Motion artifact correction in MRI using generalized projections

An algorithm that suppresses translational motion artifacts in magnetic resonance imaging (MRI) by using post processing on a standard spin-warp image is presented. It is shown that translational motion causes an additional phase factor in the detected signal and that this phase error can be removed using an iterative algorithm of generalized projections. The method has been tested using computer simulations and it successfully removed most of the artifact. The algorithm converges even in the presence of severe noise.

Motion artifact suppression: A review of post-processing techniques

Patient motion during data acquisition in magnetic resonance imaging causes artifacts in the reconstructed image, which for two-dimensional Fourier transform imaging techniques appear as blurring and ghost repetitions of the moving structures. While the problem with intra-view effects has been effectively addressed using gradient moment nulling techniques, there is no corresponding technique for inter-view effects with equal effectiveness and general applicability. A number of techniques have been proposed for correcting the inter-view effects, and these may be divided into those that minimise the corruption of the data, and those that post-process the data to restore the image. The techniques in the former category are briefly reviewed, then those in the latter category are examined in detail. These are analysed in terms of motion model, model parameter estimation, and data correction.

Segmentation of color images using spatial and color space information

One of the first steps needed to extract information from images for most machine vision applications is the segmentation of the image. We present a new segmentation algorithm for color images that combines both color space and spatial information. The algorithm is oriented to images that should exhibit clustering of the color space data, such as images of paper-based maps. The algorithm separates edge pixels from those in smooth regions and applies different segmentation algorithms to each group. The pixels nin smooth regions are used to segment the color space using a histogram analysis technique. These regions are then grown into the edge regions to classify the edge pixels. The algorithm is robust and fast, as verified by experimental results.

Suppression of slice selection axis motion artifacts in MRI

Patient motion during data acquisition in magnetic resonance imaging causes artifacts in the reconstructed image, which for two-dimensional Fourier transform imaging techniques appear as blurring and ghost repetitions of the moving structures. T. Mitsa et al. (1990) proposed a technique for suppressing artifacts from periodic motion along the slice selection axis. A different approach to the same problem is presented which is not restricted to periodic motion. The algorithm is verified using a simulated phantom and motion. It is also shown to perform well in the presence of noise and motion within the imaging plane.

A modified Gerchberg-Saxton algorithm for one-dimensional motion artifact correction in MRI

A novel method for the suppression of one-dimensional translational motion artifacts in two-dimensional Fourier transform magnetic resonance images is presented. It is shown that the motion causes an additional phase factor in the detected signal and that this phase error can be removed using a modified Gerchberg-Saxton (1972) algorithm. The major differences between this algorithm and other phase retrieval algorithms are: (1) the phase information is not totally unavailable, but in a corrupted form; and (2) the algorithm does not try to recover the entire phase information from the magnitude of the detected signal, but rather to correct the distorted phase using the average phase error. The method has been successfully tested using computer simulations. >

Iterative restoration of MR images corrupted with translational motion

Patient motion during the acquisition of magnetic resonance imaging data causes loss of resolution and ghost repetitions of the moving structures in the reconstructed image. In this paper the motion is modeled as being translational, and it is shown that this causes either the magnitude or the phase of the data to be corrupted, depending upon whether the motion is within or perpendicular to the imaging plane. The problem of restoring the image using only the corrupted data and no knowledge about the motion is addressed. The restoration problem is nonlinear in general, but is linear in two special cases. An iterative algorithm is developed that uses projections onto convex sets for magnitude retrieval and generalized projections for phase retrieval. In both cases constraint sets containing all a priori knowledge are used, and this is shown to be necessary for rapid convergence. The two algorithms may be combined to restore images corrupted by three-dimensional motion. The algorithms were verified using simulated data.

A Practical Algorithm for Structure and Motion Recovery from Long Sequence of Images

In this paper we present an algorithm for structure and motion (SM) recovery under affine projection from video sequences that is suitable for real time applications. The algorithm tracks the motion of a single structure, be it an object or the entire scene itself, allowing for any type of camera motion. This could be used for example to track the motion of a vehicle in a warehouse (single object, static camera) or for visual navigation from a moving platform (track scene from moving camera). The algorithm requires a set of features to be detected in each frame, and that at least four features are correctly matched between each three consecutive frames. Compared to previous algorithms, this novel algorithm has a lower computational cost, dynamically detects outliers and allows for previously lost features to reappear in the sequence. The algorithm has been tested on real image sequences, and compared to other algorithms we have found that our algorithm has both a smaller error and a lower computational time.

Correcting slice selection axis motion artifacts in MR imaging

Patient motion during data acquisition in magnetic resonance imaging causes artifacts in the reconstructed image, which appear as blurring and ghost repetitions of the moving structures. The acquired data are related to the image by the Fourier transform. With translational motion perpendicular to the imaging phase, the magnitude of the acquired data is corrupted. This magnitude retrieval problem can almost always be solved in theory provided the field of view is larger than the image. The restrictions on the manner in which the magnitude may be corrupted result in the problem being linear. The corruption is determined using a least squares estimate, then the data is corrected. The implementation of the algorithm allows rapid correction of images.<<ETX>>

The Eigenspace Method for Rigid Motion Recovery from less than Eight Point Correspondences

In this paper we propose a new method for the estimation of rigid motion from two monocular images when less than eight point correspondences are available. The motion parameters are found using the essential matrix. By employing previously unused constraints on essential matrix, we show that it can be estimated through the minimisation of a two-dimensional cost function defined over the space of all possible directions of translation. The new formulation is easier to understand and implement than previously proposed approaches, and has a low computational cost. The algorithm has been evaluated on synthetic data. Our experiments show that that the new method is capable of finding all solutions and that choice of initial state is not critical.