Transactions on Mass Data
Analysis of Images and Signals
Vol. 8, vol. 1 (2017) 17-30
© 2017, ibai-publishing,
ISSN: 1868-6451,
ISBN: 978-3-942952-55-2
Online ISSN 2509-9353
Verification of Hypotheses generated by Case-Based
Reasoning Object Matching
Petra Perner
Institute of Computer Vision and Applied Computer Sciences, IBaI
PSF 301114, 04251 Leipzig, Germany
pperner@ibai-institut.de, www.ibai-institut.de
Abstract. Case-based reasoning object-matching is the methods at choice when
the objects can be identified by case models. The result of the matching process
will be many hypotheses for the true shape of the objects. These hypothesizes
must be verified in a hypothesis-verification process. We review in this paper
what has been done so far and present our hypothesis-verification rules. The rules
get evaluated and the results are shown in the images and the achievements are
discussed. We consider two different hypothesis-verification rules, one is based
on set-theory and the other one is based on statistical measures. Finally, we
describe the results achieved so far and give an outlook about further work.
Keywords: Case-Based Reasoning Object-Matching, Hypothesis-Test Verification, Set Theory,
Statistical Measures
1 Introduction
Case-based object-matching [1] is the methods at choice when the objects can be
identified by case models. These case models can be learnt from the raw data by case
mining [2]. For the case-matching procedure, we need a proper similarity measure that
depends of the case model description. In our case, the case models are object contours
such as round, ellipse-like, or more fuzzy-like geometric figures. The chosen similarity
measure in this work is the cosine-similarity measure [3]. The properties of this
similarity measure have been described in detail in [3]. The case matcher takes the case
models and matches them against the objects in the image. In case the similarity
measure is high the found contour will be marked in the image. Often the matcher does
not bring out only one contour for an object, instead of the matcher fires several times
at slightly different spatial positions in the image for the same object. These multiple
matches must be evaluated after the matching in a hypothesis verification procedure.