Hierarchical labelling for integrating images and words

Abstract

This paper proposes a parallel and distributed computational model called Cellular Frame Model for hierarchical labelling of partial and global parts of images by words. The labelling is regarded as a process for integrating images and words. An objective model world is described by defining a set of labels ordered in a hierarchy by means of a Cellular Frame knowledge representation. The existence reliability of each label is determined for each location of an array in the Frame. It is determined by a state equation whose initial condition is given by an input primitive label-pattern. The state equation is of the form of local, parallel and iterative computation. The computational process described by the state equation is considered as a process of aggregation (based on input label patterns) of label knowledge introduced for the model description. In this paper, the model description of the Cellular Frame is discussed from two viewpoints: (1) label hierarchy and (2) graphical interpretation of the model. Convergence of the labelling process is ensured by the fact that the state equation always converges. In addition, several hypotheses have been ascertained by showing an example model and a simulation of the labelling process by a state equation.