Font Size: a A A

Fuzzy Active Contour Models In Image Segmentation And Change Detection

Posted on:2016-04-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:J ShiFull Text:PDF
GTID:1108330464962879Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
With the development of imaging equipment, the types of images tended to be diversification. Tthe influences of all kinds of factors may cause images have many uncertain and inaccurate factors. Complications for image segmentation include high noise, low contrast, and intensity inhomogenity. These complications make it harder to obtain accurate segmentation results. Fuzzy sets have a good ability of describing uncertainties of the image. Through the successful use of fuzzy membership degree, fuzzy segmentation methods could reserve more original image information and well describe uncertainties of the image.This thesis mainly made some efforts on studying existing problems of image segmentation methods based on fuzzy sets thoery. In order to improve the ability on the description of uncertain information in image, an in-depth study on this problem is done. Some effective methods have been proposed for actual problems of fuzzy segmentation methods. In summary, the major contributions are outlined as follows:(1)Hybrid clusteitng algorithm is considered to be a competitive method in recent years. It combines the virtues of fuzzy sets and rough sets in describing uncertainties. In order to overcome the limitations imposed by the need of manual initialization in hybrid clustering algorithms and for an automatic clustering, we design a technique to adaptive adjust weights according to the intrinsic of each dataset at each iteration. Clustering and image segmentation have a close relationship. Image segmentation can be considered as the clustering of pixels. The designed hybried clusteing algorihtm also lays the foundation for its further application in image segmentation.(2)Active contour models are usually called piecewise-constant models, since they are designed on the assumption that images are approximated by regions with piecewise-constant intensities. In real applications, due to the influences of all kinds of factors, images of such may be seldom encountered. Pixels intensities belonging to differnet regions generally have overlapping. With the introduction fo the gaussian kernel function, the improved model controls the effective scope of neighborhood window and then makes full use of image local statistical information,. With the help of image local statistical information, even the object region with little difference from the background can be well detected.(3)In order to solve the problem of intensity inhomogenity and low contract in auroal oval images, a partition-based fuzzy active contour model for auroral oval image segmentation is proposed. In real application, with the limits of imaging conditions, the intensity of background can be similar to that of the aurora region. Therefore, it is difficult to detect the complete auroara region. With the considering of image local statistical information, even the aurora oval region with little difference from the background in terms of pixels intensity can be well detected. However, the models by only considering local image information are easily trapped at local minima. To employ the local image information for dealing with intensity inhomogeneity, while avoiding getting stuck at local minima by only using the local image information. The technique which enables the contour to evolve gradually may prevent the local model getting stuck at local minima. The main idea of the technique is to limit the update area to the regions near the current contour boundaries, thus, enabling the contour to evolve gradually. Experiments on real aurora oval images have shown that the improved fuzzy active contour model can detect a clear and accurate auroral oval region.(4)With the introducting of fuzzy theory, fuzzy segmentataion methods obtain better results than “hard” partition methods. However, traditional fuzzy sets tend to capture vagueness through precise numeric membership degrees. This poses a contradiction of excessive precision in describing uncertain phenomenon. Membership degrees are more reasonable to be taken as uncertain instead of as certain in traditional fuzzy sets. To increase the ability on describing uncertain phenomenon in images, the proposed model, as an extension of traditional active contour model, generates uncertain membership degrees by using two fuzzy coefficients. Experiments on a series of images which contain large uncertainties show its effectiveness in handling uncertainties in images when compared to traditional active contour model.(5)Due to speckle noises in SAR images, the intensities of pixels belonging to changed and unchanged regions generally have overlapping. It is difficult to separate changed and unchanged regions by sharp boundaries. Interval type-2 fuzzy sets have shown their effectiveness in handling uncertainties in comparison to traditional fuzzy sets. However, the selection of fuzzy coefficients has great impact on final results and requires careful consideration in practical applications. Therefore, the interval type-2 fuzzy active contour model is employed to provide preliminary analysis of the difference image by generating intermediate change detection masks. Each realization of the intermediate change detection mask has a cost value. A genetic algorithm is employed to find the final change detection mask with the minimum cost value by evolving the realization of intermediate change detection masks. Experimental results on real synthetic aperture radar images demonstrate that change detection results obtained by the improved fuzzy active contour model exhibits less error than previous approaches.
Keywords/Search Tags:image segmentation, fuzzy theory, active contour model, change detection, auroral oval image segmentation
PDF Full Text Request
Related items