Zusammenfassung
This paper proposes an efficient scale and rotation invariant 2-D
object recognition method using Complex-Log Mapping (CLM) and Translation
Invariant Neural Network (TINN). CLM is known as very useful transform
for extracting scale and rotation invariant features. However, the
results are given in a wrap-around translated form, which requires
subsequent wrap-translation invariant recognition steps. Recently,
a new method using an augmented second order neural network (SONN)
was introduced as a solution. It requires, however, a connection
complexity O(n2) for input feature extraction which is too high to
be implemented. In this paper, we propose a method reducing the connection
complexity to O(n*log(n)) by using TINN. Experimental results show
that the recognition performance of the proposed method is almost
the same as that of SONN while its network size is significantly
reduced
Nutzer