Штучний інтелект

Анотація: This work is devoted to object search in images. To solve this problem, the method based on the use of a graph that describes the image is applied. The given graph should have such property that the individual objects in the image must comply with its subgraphs. In such case, the task of object search is limited to allocation of corresponding subgraph. To describe the images, it is suggested to use a graph constructed on the basis of the structural elements, which are identified with the help of multi-scale representation. This graph can reflect the structure of both textured and non-textured objects. To separate subgraphs, which correspond to the desired objects, the tools based on the construction of additional graph pyramid are used. Experimental verification of a given graph, and means of processing to search for textured and weakly textured objects demonstrated their high efficiency. Comparison to analogues showed significant superiority of the proposed means at searching poorly textured objects.

Ключові слова: object search, image recognition, graph comparison

Посилання: