(10) Determine whether the target is clear, Using the artificial judge, if not ideal, return to the step (5); Otherwise, complete image segmentation and finish the algorithm.
Fig.1. Original image



This value is selected too small, the number of the minimum point class to be eliminated each time small. If this value is selected too large, the target area is easily removed. The experience has shown that it is more appropriate to take value of n1 as one-tenth of the total class value m. (2) Determine the value of m
The smaller value of m is, the faster the speed of the system enhancement is, but the slower the enhancement effect is. If m≈n1, it is easy to remove the target area. The greater the value of m is, the slower the speed of the system enhancement is, but the faster the enhancement effect is. Experience has shown that: the number of categories is best able to meet each class to take M × N/30 pixels.
Determine the Parameters of the Algorithm:For the
dim and small target, the difference between the target gray
and the surrounding gray is necessarily small. In the
classification, they can be easily divided into a class. If a
target area is divided into a class in the beginning
classification, it cannot be called a dim target [17].
Therefore, this algorithm is very safe for the dim and small
Fig.3. The enhanced images of the fuzzy C-means clustering target. The main difficulty of this algorithm is to determine m
Fig.2. The enhanced images of the direct gray clustering
Experimental results
To verify the effectiveness of the algorithm, we conducted comparative experiment of the algorithm with standard C-means algorithm. Figure 1 is for the dim and small target image (the target is in the center of frame) collected by the photoelectric theodolite. The contrast of the target and background in the image is not obvious, artificial identification of the target is very difficult. Figure 2 is for the image enhancement of the direct gray clustering. Although a certain increase reached, the outline of the target boundary was vague and not completely was separated from the background. Figure 3 is for this improved algorithm. First the histogram equalization processing of the original image was carried out. Here, the number of histogram classification and mean number of clusters were taken to be 30. The value of n1 was taken to be 3. Figure 3 can be seen, target and background can already be clearly distinguished out.
and n1. Experiments showed that the value of m is close to the image histogram. If the gray of the histogram is too concentrated, the value of m should be taken smaller; the contrary, the value of m should be taken larger. The value of n1 is close to the noise of the image. If the noise is more, the value of n1 should be made larger; the contrary, the value of n1 should be taken smaller. The disadvantage of this method is that it is possible to have some damage on the background information, especially when the difference between background and the target in the gray is large. The purpose of processing the infrared images is the precise positioning for the outline of the target, so this algorithm does not affect the positioning accuracy of the image. (1) Determine the value of n1
The value of n1 determines the number of the minimum point class to be eliminated after classification each time.
Conclusion
Arming at the infrared dim and small contrast degree being low and the artificial identify of the target image being difficult, we proposed a technique based on fuzzy C-means clustering algorithm for image enhancement. A large number of the image data processing experiments showed that the image enhancement effect of the algorithm is obvious. The algorithm can enhance the gradient value of the gray in the border region between the target and background under the condition of keeping the target boundary contour not changing. It has a certain noise suppression effect. It can be applied to the enhancement processing of the very weak infrared target image.
PRZEGL D ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 3b/2012 3
A New Image Enhancement ... 暂无评价 4页 2财富值 Improving fuzzy c-...based on fuzzy c-means clustering in which some membership grades are given...
New two-dimensional fuzzy C-means clustering algorithm for image segmentation_专业资料。JCn.ot i.eho.20)58287.etSuhUnvTcni(081:8~8 DoI0.O/17—0~...
scale intervals and fix on the points, a Remote Sensing Image Enhancement Algorithm Based on Weighting Fuzzy C-Means Clustering (RSIE-WFCM) is proposed....
A New Image Enhancement ... 暂无评价 4页 2...(hard) c-means and fuzzy c-means clustering ...algorithm with a new Gaussian function based ...
Remote Sensing Image Classification Based on Improved Fuzzy c-Means_专业资料 暂无评价
which for the effect of a sensor, atmosphere etc, an remote sensing image enhancement method based on weighting fuzzy c-means clustering (RSIE-WFCM) ...
A New Image Enhancement ... 暂无评价 4页 2财富值 Improving fuzzy c-... Clustering based on gene... 2页 免费 Parallel K-Means Cluster... 6页...
的图像分割方法 ESFCM (Edge-based Semi-Fuzzy C-means Clustering) 该方法利用...Based on the inherent character that the fuzzy part in an image should be...
a 2-dimension Fuzzy C-Means (2DFCM) clustering based algorithm to segment ...[C]//International Conference on Image Processing.New York, USA: IEEE, ...
Fuzzy C-Means Clustering Course Project Presentation...? ? Image and speech enhancement Edge detection ...[PalPB97] Resubstitution errors based on the ...
热门文档