پردازش گراف هاي داده اي به وسيله مدل هاي وفقي و كاربرد آن در پردازش تصاوير

Data Graphs Processing Using Adaptive Models and Its Application in Image Processing

Recursive neural networks are one of the adoptive models that suited for processing graphs. Originally recursive neural networks are proposed to processing DAOGs in which the children of any given node are ordered. This assumption is unnecessary constraints. In this paper the weight sharing method that can relax the constraint on the ordering of children in graph processing is introduced. Also a new model of recursive neural networks to process DAG-LE is introduced. Both methods guarantee the independence of the network output with respect to the permutations of the arcs leaving from each node. In continuation an object recognition based on the last recursive model is presented. This method dose not include any priori knowledge on the object model and it is invariant under image translation and rotation.