Research On Remote Sensing Image Fuzzy Edge Detection Based On Object Cloud

The edge detection for remote sensing image, which stands for the future development direction of image edge detection, is one important way to obtain the remote sensing information and the foundation for understanding the remote sensing images. Most current algorithms for complicated remote sensing images cannot obtain a satisfying result because firstly these algorithms just take normal images as processing object and the theory form of the algorithms has not been completed; Secondly, the commonly used fuzzy algorithm takes the fuzzy sets theory as foundation and it solves the fuzzy and non-random problems in the picture. For the remote sensing image which can be one kind of random variable in some degree, the validity of the algorithm is not so good; Thirdly, as the data in remote sensing is very large and complicated, the ordinary algorithms should improve their efficiency and it becomes a realistic need to look for an efficient algorithm for remote sensing edge detection.Based on the present research situation of image edge detection, this paper takes the theory of fuzzy sets theory and cloud theory as theory foundation, and makes a research on the theory system and detection methods of fuzzy edge detection technology. This paper proposes a object-cloud based (OCFD) fuzzy edge detection method for remote sensing image based upon thorough research on the uncertainty of remote sensing image. We have made some progress in the following aspects:We have conducted a massive research on the uncertainty of the objects in the remote sensing image. According to the research, we have proposed the performance ways of different objects in the cloud space and that the object can be described in the cloud space, then we take point cloud, line cloud and plane cloud as the point like, line like and plane like objects on the earth. The data in the image space can be mapped to the cloud group through the One to Multeity model, and then it will have the digital characteristic to describe its fuzziness and randomicity and it can show the uncertain transition information. According to the characteristic of remote sensing image, we proposed the object cloud generator algorithms based on the classical cloud generator algorithms, which includes the one-dimensional cloud spatial mapped model based on the pixel's gray level characteristic, the two-dimensional cloud spatial mapped model based on the gradient characteristic of the image in two directions. In addition, we have analyzed the spectrum information and data characteristic of multi-spectrum remote sensing image, generalized the one-dimensional cloud space to multi-dimensional space according to the multi-spectra image and established the multi-dimensional cloud spatial mapped model for multi-spectra image. The result of experiments shows that the fuzziness and randomicity of the image object can be performed well by the digital characteristic of the object cloud in cloud space.We have proposed some methods to obtain the one-dimensional boundary cloud and multi-dimensional boundary cloud. The digital characteristic of boundary cloud can be obtained by some operations in the cloud's neighboring region and the correlations between pixels are made good use of. This method is like the smoothing method, so it can resist the noise at some point; On the other hand, the entropy and the hyper entropy of the boundary cloud can be obtained by calculation. The entropies and hyper entropies of the symmetrical clouds have some strong relativity which reflects that the image randomicity affects the expectation and criterion difference of the boundary cloud.We have proposed the transition region extraction algorithm based on boundary cloud. Using the entropy and the hyper entropy of the boundary cloud, we can obtain the left and right gate-limited value without worrying that the minimum gray value can be larger than the maximum gray value as in the traditional method based on the transition region; Having small calculation cost, this algorithm is fast and simple and can resist the noise at some point. We have proposed the fuzzy characteristic plane of the digital characteristic of boundary cloud and the method to create it. We have also defined each element in the fuzzy matrix as a gathering of subordination values, which shows the subjection level of each pixel in an image is not a exact value, but a kind of probability distribution that shows the influence the randomicity gives to the result of image detection at some point. The image edge detection algorithm based on maximum fuzzy entropy has been provided at last. This algorithm has discussed the edge characteristic according to the neighboring region coherence estimate defined by the theory of maximum fuzzy entropy and considered the relativity among the pixel and its neighboring region; the algorithm makes use of the data of the fuzzy characteristic matrix to calculate the fuzzy entropy, also takes the random effect to the calculation of the entropy, and seeks the optimal solution in the membership set in which the maximum fuzzy entropy obeys a certain possibility distribution; In order to make best use of the information in an image, we have provided the edge points detection principles and methods in the restraint of the frame coherence estimate and the direction coherence estimate. This algorithm considered the characteristics of the edge and used the image information in maxium degree in edge detection. Experiments testified that this algorithm could improve the edge detection result in some point.