Human Object Classification Using Dual Tree Complex Wavelet Transform and Zernike Moment - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 94

5 Experimental R Results

This section provides quan
other state-of-the-art metho
and Zheng [9], Renno et al.
object classification results
son of the results obtained
has been done in terms of c
tive rate (recall), and pred
human object classification
shown some human and no
Fig. 2. By observing the hu
one can observe that both f
experiments.

ntitative experimental results of the proposed method
ods proposed by Khare et al. [2], Dalal and Triggs [7],
. [14], and Chen et al. [15]. We have also evaluated hum
by using discrete wavelet transform as a feature. Comp
by the proposed method and other state-of-the-art meth
confusion matrix, average classification accuracy true po
dicted positive rate (precision). The proposed method
n has been tested on INRIA person dataset [29]. We h
n-human representative images of INRIA person datase
uman object image of INRIA person dataset from Fig. 2
frontal as well as side view of human object were taken

and
Lu
man
pari-
hod
osi-
for
have
et in
2(a),
n for

(a)

(b)
es with human and non-human objects of INRIA dataset

Fig. 2. Sample image

For experimentation, we
ment, (ii). Only Dual tree c
and (iii). Combination of D
nike moment coefficients. W
of Dual tree complex wavel
for the proposed method ap
given in Table 1, 2 and 3 r

e have taken three types of features: (i). only Zernike m
complex wavelet transform coefficients (Khare et al. [
Dual tree complex wavelet transform coefficients and Z
We have evaluated the proposed method for multiple lev
let transform coefficients (L - 1,2,….,7). Confusion ma
pplied by using these three above mentioned feature set
respectively. Just to compare performance of the propo

mo-
[2]),
Zer-
vels
atrix
are
osed