Resolution analysis for Gradient Direction Matching of object model edges to overhead images

https://doi.org/10.1016/j.cviu.2008.09.002Get rights and content

Abstract

The problem of computer-assisted broad area search for specific objects of interest in overhead images is considered. To this end, we present a novel efficient Gradient Direction Matching (GDM) algorithm that matches gradient directions associated with object edges to pixel gradient directions (as opposed to image edges, which are less reliable). GDM seamlessly integrates information associated with pixel location and orientation in such a way that the FFT can be exploited for computational efficiency, and it inherently rejects background clutter. The effects of spatial resolution on GDM statistical performance are studied empirically with the goal of gaining insight into how far GDM computational cost can be reduced before matching performance becomes too severely compromised.

Section snippets

Background and introduction

This paper addresses the problem of broad area search for specific objects or families of related objects in sets of overhead images (aerial or satellite) with broad area coverage. This is a problem of major importance to human analysts and photo-interpreters in government and military services responsible for quickly analyzing enormous volumes of overhead imagery. Computer assistance is required because human analysts are only able to manually search large volumes of overhead imagery at a

Gradient direction fields

GDM requires a field of gradient directions for image pixels and gradient directions for object edges projected onto the image. Gradient directions are estimated differently in each case.

Gradient Direction Matching

For an 8-bit image u with w columns of pixels (the width) and h rows of pixels (the height), the wxh array θ of pixel gradient directions θ(c, r) can be estimated using (1). A corresponding wxh binary array of thresholded pixel gradient magnitudes a(c, r) can be generated by setting a(c, r) = 1 only when the gradient magnitude A(c, r) in (1) is greater than some minimum perception threshold A0 on 8-bit gray-value contrast (e.g., A0 = 8).

Next, suppose the physical model of some object to be detected

GDM performance at variable matching resolutions

This Section addresses two fundamental issues related to GDM efficiency and statistical performance, namely (1) how small the object model projections can be, and (2) how many object rotations need to be matched. An understanding of these issues allows image spatial resolution and the number of object rotations to be chosen prior to model matching so as to maximize the pixel throughput rate while maintaining good matching performance.

Suppose N rotations of an object are to be matched to an

Summary and conclusions

A novel Gradient Direction Matching (GDM) algorithm that matches physical models of objects to overhead images with broad area coverage has been presented. It was designed to handle variations in scene clutter and image acquisition conditions. As a hybrid algorithm, GDM incorporates philosophies consistent with both edge-based matching and image-based matching. It matches orientations of object model edges on the one hand to image pixel gradient directions (as opposed to image edges, which are

Acknowledgements

The authors thank George Weinert for implementing the image thumbnail-based query interface and running scripted experiments. Thanks also to Chuck Grant for developing and implementing the algorithms that project object model edges onto overhead images.

References (31)

  • L. Brown

    A survey of image registration techniques

    ACM Comput. Surveys

    (1992)
  • J. Canny

    Computational approach to edge detection

    IEEE Trans. PAMI

    (1986)
  • F. DeCastro et al.

    Registration of Translated and Rotated Images Using Finite Fourier Transforms

    IEEE Trans. PAMI

    (1987)
  • W. Eppler, D. Paglieroni, S. Petersen, M. Louie, Fast normalized cross-correlation of complex gradients for...
  • W. Eppler, B. Trusso, System for Registering Site Models to Gray-Scale Images”, unpublished white paper, NIMA Contract...
  • Cited by (7)

    • Gradient orientation pattern matching with the Hamming distance

      2014, Pattern Recognition
      Citation Excerpt :

      Most conventional pattern matching techniques may be classified into two categories: one is based on image intensities, including colors, and the other is based on gradients or edges. Since image intensities and gradients are dependent on lighting conditions, these traditional approaches often fail to perform matching correctly under irregular lighting conditions [24]. To overcome this problem, the use of gradient orientation has drawn attention in recent years, as it is known to be a robust image feature to varying illuminations [3,7,22].

    • Improved direction estimation for di Zenzo's multichannel image gradient operator

      2012, Pattern Recognition
      Citation Excerpt :

      Gradient [1] has been widely used in image processing and computer vision in applications such as edge detection [2–6], image segmentation [7,8], corner detection [9], image fusion [10], image recognition [11], face detection [12], and object tracking [8,13].

    • Generating fuzzy edge images from gradient magnitudes

      2011, Computer Vision and Image Understanding
      Citation Excerpt :

      However, most of the methods are based on the representation of the intensity change by means of a vector, usually called gradient [3]. This observation applies to the methods of Canny [4,5], Sobel and Feldman [6] and Prewitt [7], but also to some more recent proposals [8–12]. The magnitude of the gradient plays a major role in further processing steps.

    • Prejudgement of hot-images' damage based on gradient direction matching method in high power laser system

      2017, Proceedings of SPIE - The International Society for Optical Engineering
    View all citing articles on Scopus

    This work was performed under the auspices of the U.S. Department of Energy by the University of California, Lawrence Livermore National Laboratory under Contract No. W-7405-Eng-48 (UCRL-JRNL-217038).

    View full text