Human Object Classification Using Dual Tree Complex Wavelet Transform and Zernike Moment - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 89

2 Feature Selection

Castleman [16] defined feature as, “A feature is a function of one or more measure-
ment computed so that it quantifies some significant characteristics of object”. In any
object classification algorithm, selection of appropriate feature is very important. If
corrected feature is selected for classification algorithm then performance of the clas-
sifier will increase. Use of single type of feature may not be sufficient for solving
human object classification problem, because single type of feature is not rich enough
for representation of different shades of human objects. Combining multiple type of
features can enhance the robustness and classification accuracy of the classifier for
human object classification. When we use combination of two or more features, some
of the features are more informative than others for a particular case, therefore chanc-
es of correct classification will be high. In the proposed work for human object classi-
fication, we have taken combination of two different features - Dual tree complex
wavelet transform (DTCWT) and Zernike moment (ZM). A brief description of these
two features and why they are useful for human object classification are described in
subsection 2.1 and 2.2 respectively.

2.1 Dual Tree Complex Wavelet Transform

Real valued wavelet transform suffers from two major shortcomings - lack of shift
sensitivity and lack of strong edge detection. Kingsbury et al. [17,18] proposed a solu-
tion to overcome these problems in form of use of Dual Tree Complex Wavelet
Transform (DT-CWT). DT-CWT is shift invariant in nature and gives strong edge
information for filtering multidimensional signals. Dual tree complex wavelet trans-
form have following properties as described in [17, 18].

- Approximate shift invariance.

- Good directional selectivity in 2-D with Gabor like filters (Also true for m-

dimensional ).

- Perfect reconstruction using short linear phase filters.

- Limited Redundancy, independent of the number of scales ( 2:1 for 1-

Dimensional and 2 n :1 for n-Dimensional ).

Efficient order N computation - only double of the simple DWT for 1-D ( 2 n
times for n-dimensional ).

-

Human object classification is a problem where the objects may present in translat-
ed as well as rotated form among different scenes. Therefore first three properties of
dual tree complex wavelet transform will be useful for classification of human object.
For implementation of DT-CWT, Kingsbury [18] analyzed that the approximate
shift invariance with a real DWT can be achieved by doubling the sampling rate at
each level of the wavelet decomposition tree. For this work, the samples must be
evenly spaced. One approach to double the sampling rate in Tree a of Fig. 1(a) is by
eliminating the down sampling by 2 after the level 1 filters, H 0a and H 1a . This is
equivalent to two parallel fully decimated tree a and b in Fig. 1(a), provided that the
decays of H 0b and H 1b are one sample offset from H 0a and H 1a which ensure that the