Abstract
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
Users
Please
log in to take part in the discussion (add own reviews or comments).