Feature Extraction Based On Image Potential Energy

Image processing technology provides a vast space for computer vision. Target recognition is an important and indispensable field in computer vision system. Research feature extraction of target images and target detection has important theoretical significance and practical value. It can give a variety of theoretical basis and methods for computer vision system. It also can provide computer-aided support in the development of national economy, such as in national defense, medical, aerospace, industrial production, agricultural production, emergency rescue and disaster relief and so on. The traditional target detection technology use overall feature or local feature for target recognition, but it is not fit for the request of the new target detection field. Image potential theory is using target image potential energy to describe object feature, which can give the overall feature and local feature of the target at the same time. The two kind of features both are the reliable judgment basis of target recognition, and the features can mutual independence and interdependent.In the study of image feature extraction, according to the image’s physical properties and visual effect, the potential energy method was used to mark target feature by relative positions based pixels, and then using image projection and the theory of physics, this paper the firstly proposes the concept of image potential energy. Which finished some basis work, such as definition, derivation, analyzing, simulate experiments and so on. In the image analysis, according to study the characteristics of image, the classification, put forward the one-dimension potential, two-dimensional potential energy, standard potential energy, skeleton potential energy and contour potential energy. Through the theoretical analysis and the experiment of animals, number and model respectively, summarizes different kinds of target fit for their only potential energy, and in this time the system has the best effect.In target modeling and target detection study, the image modeling method based on potential energy and the potential energy distance based on vector distance detection methods are mentioned. The potential energy vector is used to store image feature and used to recognize target, so we need creating a good data base within potential energy vector for target recognition system. By using Euclidean distance method to calculate the potential energy distance can get the level of similarity of two sample, so as to achieve the target detection function. In target detection simulated experiments,the better detection effect was received according to various font numeral symbols similarities and differences calculate the one-dimensional potential energy. In the plane model testing, using sample median agglutination to detect the different range model can avoid the dynamic target detection problem.Paper mentioned the new theory and analysis, definition, application and simulation experiments within different targets, and got a satisfactory experiment result.