![](bg1.png)
An Anisotropic and Asymmetric Causal Filtering Based Corner
Detection Method
Ghulam Sakhi Shokouh
a
, Philippe Montesinos
b
and Baptiste Magnier
c
EuroMov Digital Health in Motion, Univ. Montpellier, IMT Mines Ales, Ales, France
Keywords:
Causal Filter, Anisotropic Filtering, Asymmetric Diffusion, Corner Detection.
Abstract:
An asymmetric-anisotropic causal diffusion filtering-based curvature operator is proposed in this communica-
tion. The new corner operator produces optimal results on small structures, such as, corners at pixel level and
also sub-pixel level precision. Meanwhile, this method is robust against noises due to its asymmetric diffusion
scheme. Experiments have been performed on a set of both synthetic and real images. The obtained results are
promising and better without any ambiguity as compared with the two referenced corner operators, namely
Kitchen Rosenfeld and Harris corner detector.
1 INTRODUCTION
The importance and interest in keypoint detection (i.e,
corner or junction as a stable interest point) in a digital
image lies notably in its application in image match-
ing, tracking, motion estimation, panoramic stitching,
object recognition, and 3D reconstruction (Schmid
et al., 2010). The reason for the corner detection’s
wide range of applications is that the corner is eas-
ier to localize than other low-level features such as
edges or lines, particularly taking into consideration
the correspondence problems (e.g., aperture problem
in matching). There are many corner detection tech-
niques based on classic handcrafted (Shokouh et al.,
2023) and deep learning methods (Wang et al., 2019).
Deep learning-based techniques are more automatic,
however, considering the accuracy and precision for
the detection of small structures, such as corners,
keypoints, etc., they do not generally present higher
performance (highly depends on the dataset quality,
and annotation which is not easy for small struc-
tures). We argue that handcrafted techniques are still
widely used, particularly for optimization purposes,
either independently or integrated into the prepro-
cessing or post-processing stages of machine learn-
ing based higher-level computer vision applications
(Junfeng et al., 2022). One of the example of com-
puter vision application that its performance directly
a
https://orcid.org/0000−0003−2561−7317
b
https://orcid.org/0000−0003−3741−8702
c
https://orcid.org/0000−0003−3458−0552
depends on the precision and accuracy of keypoint de-
tection is 3D reconstruction. Additionally, among the
classic handcrafted corner detection techniques, the
two corner detection operators Kitchen and Rosen-
feld (Kitchen and Rosenfeld, 1982), and Harris (Har-
ris and Stephens, 1988), are the main method used
for the comparison and benchmarking. Moreover,
Causal filtering has proven its efficiency in many
segmentation domains, such as edge or line detec-
tion. In this contribution, we are presenting a new
segmentation method for corner detection based on
asymmetric anisotropic diffusion filtering. The ba-
sic idea is inspired from a curvature-like operator
similar to the Kitchen-Rosenfeld operator, but imple-
mented through an asymmetric diffusion scheme us-
ing an anisotropic causal filter. Finally, we have com-
pared the experimental result of our operator with the
Kitchen and Rosen- feld, and Harris, the visual re-
sult presented higher precision for both pixel level and
sub-pixel level. The structure of this paper consists
of related works in the subsequent section, followed
by the proposed methods and the obtained result, and
eventually, the conclusion is presented.
2 RELATED WORKS
Considering a curve traced on the image plane, the
curvature is defined as
dθ
ds
where θ is the tangent to the
curve and s the curvilinear coordinate along the curve.
As an image, I(x, y) is a Cartesian parametrized sur-
92
Shokouh, G., Montesinos, P. and Magnier, B.
An Anisotropic and Asymmetric Causal Filtering Based Corner Detection Method.
DOI: 10.5220/0011855600003497
In Proceedings of the 3rd International Conference on Image Processing and Vision Engineering (IMPROVE 2023), pages 92-99
ISBN: 978-989-758-642-2; ISSN: 2795-4943
Copyright
c
2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)