Abstract
Paper describes an investigation about the problem of image recognition in computer vision based on a set of structural SURF-features. Self-organization process is proposed to be performed in space of structural features with a goal to increase recognition process performance. Kohonen neural network is used as self-organization method. The object of research is the method of similarity calculations and models of intelligent data processing in the new feature space. The subject of research is the systematization and grouping of sets of structural features of visual objects. Goal of a paper is to construct structural recognition method based on input data as a set of cluster structural features obtained as a result of self-organization. The objectives of the research are the investigation of the features and analysis of models to calculate clusters of features, the construction of the modified measures of structural similarity, the experimental evaluation of the recognition quality for different ways of descriptions comparison in the application-based visual image database. Construction of an image recognition method based on etalon descriptions as a cluster was proposed, recognition is based on the classification of the structural features of an object in cluster space with further calculation and optimization of the similarity vector descriptions. Experimental investigations and simulations of the proposed recognition method on the test image set with the use of SURF characteristic features were performed. Performance boost and efficiency of the method were confirmed, estimation of recognition quality for different processing options was performed.
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call