Sar Image Segmentation Based On Evolutionary Computation

Evolutionary Computation is a simulation of biological evolution and mechanisms of adaptive artificial intelligence technology to solve problems, is a type of random search technology, which is the simulation of the population learning process. Each individual is a search space of the problem. In image processing, such as feature extraction, image segmentation, and so there are some errors inevitably, the error would affect the results of image processing, how to minimize these errors is an important practical requirement to make computer vision practical. Evolutionary Computation in pattern recognition, image restoration, image edge feature extraction, image segmentation, etc. has been applied to study and become a hot spot at present. This dissertation centers around the application of evolutionary computation of image segmentation for a number of researches, The main contributions can be summarized as follows:Evolutionary computation often used as a optimization method in image segmentation, and how to encode the image is the key step. Existing encoding method is generally more complicated, or there is a lot of parameters, In this dissertation, a simple method of coding was used that is coding to the relationship of adjacent regions, but this encoding method has an important flaw is that the relationship will lead to pass through the merger of error, this dissertation for this designed the operation of splitting, effectively prevent the defects. Algorithm as follows:first get the G-image which response to the region homogeneity, And then had a rough image segmentation using region growing produce initial region, On this basis, define the Affinity function, clonal selection algorithm was used to optimize. On the real SAR image segmentation experiments carried out to demonstrate the effectiveness of the algorithm. Finally, on the rich in texture information SAR image, we used nonsubsampled contourlet transform to extract features, Using the above-mentioned algorithm for the experiment and get the good result.Clustering as a classical algorithm was widely used in image segmentation, but due to lack of space between pixels on the location information into account, making algorithm used in the SAR, such as division of strong noise of the image is difficult to obtain good results. In order to reduce the impact of noise on the results, need to join the spatial information of the image. We used game of life evolutionary thinking proposed by John Horton Conway. In order to use the spatial information of image into the process of evolution, Algorithm gives the two constraints:population constraints and neighborhood constraints, population constraints reflect that population characteristics have an impact on the pixel, neighborhood constraints reflect that Spatial Information have an impact on the image pixels, the two constraints together to guide the process of evolution. Respectively on the synthetic noise image and real SAR image segmentation experiments carried out to demonstrate the effectiveness of the algorithm. Finally, the population constraints and the neighborhood constraints operator were analyzed.