Human Object Classification Using Dual Tree Complex Wavelet Transform and Zernike Moment - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 99

Table 6. ( continued )

Average
Classification
Accuracy
(%)

TPR
(Recall)
(%)

PPR
(Precision)
(%)

Methods Name

The Proposed method with combination of
DTCWT (Level-4) and Zernike moment as a
feature

96.00 95.05 95.50

The Proposed method with combination of
DTCWT (Level-5) and Zernike moment as a
feature

97.00 97.00 97.00

The Proposed method with combination of
DTCWT (Level-6) and Zernike moment as a
feature

99.00 98.02 98.50

The Proposed method with combination of
DTCWT (Level-7) and Zernike moment as a
feature

100.00 98.04 99.00

DWT (Level-1) as a feature 85.00 73.91 77.50

DWT (Level-2) as a feature 85.00 73.91 77.50

DWT (Level-3) as a feature 89.00 76.92 81.00

DWT (Level-4) as a feature 91.00 81.25 85.00

DWT (Level-5) as a feature 93.00 82.30 86.60

DWT (Level-6) as a feature 95.00 89.62 92.00

DWT (Level-7) as a feature 95.00 89.62 92.00

Method Proposed by Dalal and Triggs [7] 96.00 92.31 94.00

Method Proposed by Lu and Zheng [9] 90.00 75.00 80.00

Method Proposed by Renno et al. [14] 89.00 74.17 79.00

Method Proposed by Chen et al. [15] 98.00 96.08 97.00

From Table 6, one can observe that the proposed method with combination of Dual
tree complex wavelet transform coefficients and Zernike moment as a feature set
gives better performance at higher levels of Dual tree complex wavelet transform in
comparison to other state-of-the-art methods [7,9,14,15], and multilevel discrete
wavelet transform, as feature, for human object classification in terms of three
different quantitative performance measures -Average Classification Accuracy, TPR,
and PPR.

6 Conclusions

In this paper, our goal is to develop and demonstrate a new method for human object
classification in real scenes using combination of two feature set namely - Dual tree
complex wavelet transform coefficients and Zernike moment. Dual tree complex
wavelet transform is advantageous over real valued wavelet transform in terms of
better edge representation and shift invariant property. Zernike moment is translation
and rotation invariant. The proposed method for human object classification is an
extension of work proposed by Khare et al. [2]. The proposed method first trains
SVM classifier by using selected feature set, then classify new object data (testing