50 
the complete information and more obvious feature of foreign 
fibers may be useful for separating them from cotton. 
4. IMAGE FUSION ALGORITHM 
The goal of image fusion is to integrate complete information 
from multiwavelength image data such that the new image are 
more suitable for the purpose of human visual perception and 
computer processing task such as segmentation and object 
extraction can be improved. Many image fusion methods have 
been proposed recently, however most methods such as HIS 
[11], PCA [12], and HPF [13] distorted the original 
information of the object. The wavelet-based approaches 
preserve the spectral characteristic of the multiwavelength 
images better than the standard PCA and HIS methods [13, 14, 
15]. 
The wavelet-based image fusion methods can be performed in 
2 ways as following: 
First, the two images, 405nm and 850nm wavelength images, 
have to be registered and the wavelet transforms of images are 
computed. The larger absolute transform values in this bands 
correspond to sharper brightness changes and thus to the 
salient features in the image such as edges, lines, and region 
boundaries. Therefore a good integration rule is to select the 
larger of the two wavelet coefficients at each point. 
Second, a composite image is constructed by performing an 
inverse wavelet transform based on the combined transform 
coefficients. Fig.l shows the schematic diagram of the 
structure of the proposed image fusion scheme. This selection 
scheme helps to ensure that the dominant features are 
incorporated as complete as possible into the new images. 
Fig. 1 the schematic diagram of the structure of the image 
fusion scheme 
In the fusion scheme proposed above, an area-based selection 
rule is used. The images are first decomposed into a gradient 
pyramids, the variance of each image patch over a 3X3 or 
5X5 windows is computed as an activity measure associated 
with the pixel centered in the window. If the activity measure 
at the corresponding location are closed to each other, the 
average of the two is considered as new value, other wise the 
large value is chose. 
5. EXPERIMENT OF IMAGE FUSION 
The multiwavelength imaging system consisted of 
multispectral CCD camera, an illumination chamber, and a 
computer equipped with a frame grabber. A darkened 
chamber was built with a box with a round-open inlet on the 
top, through which the CCD camera was mounted facing 
downwards. The illumination was provided using a pair of 
LED array light source respectively at the band of 405nm, 
850nm. The dual light are mounted 40cm apart and were 
convened with plastic light diffused, 50cm from the object, 
which provide evenly distributed illumination to the cotton. 
At the same time, this two 50X50 LED array light source 
were positioned bilaterally at 450 angles to provided balanced 
area illumination to the cotton. The scheme of 
multiwavelength imaging system was given in Fig.2. 
Fig.2 The scheme of multiwavelength imaging system 
Fig.3 shows representative images of six types of foreign 
fibers on the surface of cotton, which are respectively visible 
region image, 405nm wavelength image, 850nm wavelength 
image and fused image of 405nm and 850nm wavelength. 
Fig.3 shows that, in general, cotton fibers have almost same 
reflectance at the visible band as the foreign fibers have. Then, 
visual contrast between cotton and foreign fibers in the visible 
region image is not enough to detect them. However, the 
cotton shows higher reflectance in the 405nm and 850nm 
wavelength compared to the foreign fibers, except for bristle 
and plastic, which have higher reflectance. Among the foreign 
fibers, knitting, jute, hair and wool have the lowest reflectance, 
but bristle and plastic have higher reflectance than the others. 
Clearly, wool, bristle, jute, and knitting are obvious in the 
image at the band of 405nm, while hair and plastic are 
discerned in the image at the band of 850nm. From the image 
of visible region, it is difficult to identify the image feature of 
any type of foreign fiber. However, we can find the six types 
of foreign fibers more clearly in the fused image of the two 
bands, 405nm and 850nm. Furthermore, better visual contrast 
between cotton and hair fibers is show in the fused image than 
other images. 
The result indicates that the foreign fibers may not be easily 
detected with the use of individual band or visible region. 
While, the fusion image at OOP could be useful for detecting 
these foreign fibers, which are almost invisible from cotton. 
(c) 
Fig-3 compari: 
Thus, the combinati 
enabled the differen 
best visual contrast 1 
in the fused image o 
method using mul