(لا توجد تقييمات)
Loading...بحث بعنوان
Experimental Comparison of Correlation Techniques
Jérôme Martin and James L. Crowley
IMAG-LIFIA, 46 Av. Félix Viallet, 38031 Grenoble, France
Abstract. This paper presents an experimental comparisons of different forms of SSD and
normalised cross correlation of image neighborhoods. Signal detection theory is used as a
framework for analysis of correlation techniques. A sum of squared difference (SSD) of two
image neighborhoods is shown to provide an optimal matching measure for tracking and
registration. Correlation of the image, its gradient magnitude or its Laplacian are discussed. The
relations between SSD and Cross Correlation are demonstrated, and different normalisation
techniques are described. An experimental comparison is made of SSD, Normalized Cross
Correlation, and Zero-Mean normalised cross correlation, in the presence of changes in light
level, additive Gaussian noise, salt and pepper noise, and changes in the mask size.
Keywords. Image Description, Matching, Correlation, Normalisation.
Conclusions
The appropriate choice of basis set and normalisation are
determined by the task. None-the-less, the analysis and
experiments permit us to make the following
observations. In general, SSD provides a more stable
result than NCC or ZNCC. For the sources of noise
tests, the SSD of the gradient usually provides the most
stable result. When SSD must be replaced by a
correlation, it is generally preferable to use the NCC of
the gradient. In general, NCC seems to provide a more
stable detection than ZNCC (and is also less costly)
although, we feel that this last conclusion must be the
subject of further experiments. Furthurmore, the gradient
provides a more stable result than the Laplacian.
كلمة سر فك الضغط : books-world.net
The Unzip Password : books-world.net
تحميل
يجب عليك التسجيل في الموقع لكي تتمكن من التحميل
تسجيل | تسجيل الدخول
