Colourhistogram



II. TEXTURE FEATURE EXTRACTION IN CBIR

An overview of the proposed CBIR system is illustrated

in Fig. 1. The proposed algorithm, Label Wavelet Transform

(LWT), is based on color image segmentation [1], and it is

an extension of DWT-based texture feature extraction method.

The 2-D DWT is computed by applying separable filter banks

to the gray level images. The detail images Dn,1, Dn,2,

and Dn,3 are obtained by band-pass filtering in a specific

direction, and they can be categorized into three frequency

bands: HL, LH, HH band, respectively. Each band contains

different directional information at scale n. The texture feature

is extracted from the variance (ó2

n,i) of the coefficients cn,i of

the detail image Dn,1, Dn,2, and Dn,3 at different scale n.To

represent the texture feature of an image q, the texture feature

vector of DWT is defined as [2]:

TDWT (q) = [ó2

1,1, ó2

1,2, ó2

1,3, ..., ó2N

max,3], (1)

where Nmax denotes the largest scale. In this work, Nmax