Scene analysis using an integrated composite neural oscillatory elastic graph matching model

摘要

Scene analysis is so far one of the most important topics in machine vision. In this paper, we present a neural oscillatory model integrated with an elastic graph dynamic link model to provide an automatic means of processing color images. The system involves: (1) multi-frequency bands feature extraction scheme using Gabor filters, (2) automatic figures-ground object segmentation using a composite neural oscillatory model, and (3) object matching using an elastic graph dynamic link model. Using an image gallery of over 3000 color objects, with the recognition of 6000 different scenes, our model shows an average recognition rate of over 95%. For occluded objects in cluttered scenes, the model can still maintain a promising recognition rate of over 87%. Compared with that of the contemporary scene analysis models of gray-level images based on a coupled oscillatory network, the proposed model provides an efficient solution for color images using the composite neural oscillatory model (CNOM). Coupled with the elastic graph dynamic link model (EGDLM), the object recognition process takes less than 35s on average to complete, which is quite promising in many applications.