LOCALIZATION OF GRAPHICAL OBJECTS IN RASTER IMAGE BY MEANS OF ANGULAR PATTERN SIMILARITY

The proposed localization method for graphical objects of scene image is invariant to the illuminance conditions, scale, offset and rotation of the objects. The method is based on a vector representation of scene image and searched object. Model for building of object pattern and algorithm of its recognition are represented. Рattern creation and recognition algorithm use the approach based on the analysis of evidences. Localization reduces to the verification of hypotheses about object location on real image by means of presence of invariant elements of the graphical image of the object. The image description is created in a vector form after the raster image vectorization step is completed. The transition to a vector form provides invariance to scene illumination conditions. Invariance to scale and turns is achieved by using the principle of angular similarity with the accumulation of evidences of the desired graphic image. The number of accumulated evidences can serve as an estimation of the level of result reliability of the localization problem. The paper presents experimental results for real images.

Authors: A.Yu.Dorogov

Direction: Informatics and Computer Technologies

Keywords: Invariant, graphical object, scene image, localization, angular similarity, vectorization of images, accumulation of evidences


View full article