six types of foreign fibers was calculated by the following 
formula, 
A 2=7-ZISc-g,| (5) 
o , =1 
where / = 1,2 6, and g ii g c respectively represented 
gray level of six types of foreign fibers and cotton. Table 2 
shows Ag value in different images. And it is clearly that the 
Ag value of fusion image is maximal. 
image 
Ag value 
visible region image 
18.3528 
405nm wavelength image 
24.7381 
850nm wavelength image 
22.0231 
Fusion image 
30.1598 
Table 2 ^ value in 4 different images 
In order to separate these foreign fibers from the background of 
cotton, an image segmentation method based on local contrast 
variations rather than using a predefined threshold was 
developed. Thus, the sorting of foreign fibers could be 
conducted using a hierarchical segmentation method. This 
would consist first in segmenting all the foreign fibers and then 
identifying each type of foreign fibers by means of dedicated 
image analysis algorithms. Before the image segmentation 
processing, image enhancement techniques were used for the 
fusion image so that the image features of foreign fibers are 
more obvious. Single-point processing method based on the 
intensity of single pixel was applied to enhance the image 
contrast of fusion image. 
The final step in the image processing was to use a dynamic 
thresholding process to set all pixels of foreign fibers to zero, 
from then on, only the zero pixel in each image were 
considered. After the dynamic thresholding process, all the 
foreign fibers were clearly identified in the binary image. Fig.4 
shows the enhancement image and binary image of foreign 
fibers. 
Fig.4 image segmentation processing: (a) enhancement image, (b) binary image 
After the image segmentation processing, the six types of 
foreign fibers can be identified as shown in the far right images 
in Fig.4. In these binary images, all the foreign fibers are 
clearly identified and can be easily separated from the cotton. 6 
6. CONCLUSION 
The study was conducted to determine the feasibility of a novel 
multiwavelength imaging system for detecting foreign fibers. A 
spectral imaging experiment was set up to select the most 
suited wavelength bands to sort foreign fibers. In particular, the 
separability of six types foreign fibers including hair, knitting, 
jute, plastic, wool, and bristle was studied. 
Through the experiment analysis, the OOP including optimal 
wavelength and optimal light energy for each specific type of 
foreign fibers was determined. Wavelengths were selected by 
gray level discrimination analysis of spectral image data. The 
wavelength 405nm and 850nm were chosen from the seven 
bands for use in the imaging system. An image fusion 
algorithm based on wavelet-transform for enhancing the image 
feature and acquiring complete information is proposed also in 
this paper. Then, image segmentation algorithms based on local 
contrast variations was considered when attempting to separate 
them from the cotton. This result suggests that use of 
multiwavelength imaging technique for detection of foreign 
fibers in cotton in a commercial setting may be feasible. 
REFERENCES 
[1] Michael C G, “Agriculture marketing services-the 
classification of cotton,” Cotton Ginning Engineering 
Conference, Washington, USA, pp.92-96, 1999 
[2] H M Strolz, 
International Cott 
67,2000. 
[3] P Tantaswadi, J 
automated visual in 
using color Isodisc 
Processing, vol.27, 
[4] Tae Jin Kang , 
the trash and colo 
neural network,” Te 
[5] B Xu, C Fang a 
cotton trash and ci 
no. 12, pp.881-890, 
[6] D veit, J bergn 
tool to improve m 
Clothing Sci and Te 
[7] J.S Church, J. 
contamination in 
spectroscopy, vol.2i 
[8] Wu RongHi 
identification of dif 
of Forensic Medicir 
[9] Ajay. Pai, 
contamination via 
IEEE Trans on Im 
2002