出版社：Ivan Kozhedub Kharkiv National Air Force University
摘要：Increase of the amount of intelligence processing results in high time costs. Therefore,it becomes very important to auto-mate the process of selecting objects in images. That will increase the efficiency of solving intelligence problems. The main tasks that in general significantly affect the quality of the decryption process are the tasks of detection and identification of single and complex intelligence objects in the image. The main problem when converting a bitmap to a vector format is to achieve 100% similarity between the original and final images. The criteria for segmented image quality have been defined. They are the ho-mogeneity and dissimilarity of neighboring regions;the smoothness of their borders,the small number of small “holes” within the region,the boundaries of each segment should be simple,spatially accurate. The possibility of using the existing methods of constructing a segment map has been analyzed. At it the low tone saturation of space and aerial images and the presence of a large number of textured objects were taken into account. Since the range of possible color values of pixels in the textured region varies widely,the usage of decryption features based on the color characteristics of the analyzed area does not ensure accept-able decryption results. Decrement of the range of allowable values eliminates most of the errors of the first kind,but propor-tionally increases the number of second kind errors. When using the texture features of an area as a homogeneity criterion,the exact similarity of the vector image of the object is not achieved,and the closeness of the texture kernel sizes of different objects leads to first and second kind errors. The article suggests usage of the constructed field of fractal dimensions in order to deter-mine the boundaries of different segments,in which there is an impulse change of fractal dimension values.
其他摘要：Першочерговим етапом,що визначає якість автоматичної обробки і результату дешифрування в цілому,є етап створення сегментної карти зображення,який повинен забезпечити повну подібність між вихідним і кінцевим зображенням. Визначені критерії якості сегментов
关键词：image segmentation;vectorization;color space clustering;textural features;contour filters;fractal dimension field.
