A SIFT based object recognition using contextual information

While automated object recognition in natural scenes has been studied for long times, it still remains a challenging problem in machine vision, image processing and analysis. Object recognition can be considered as the problem of classifying a given object into a number of classes. The problem of object recognition involves a several critical challenges such as scaling, rotation, distortion, illumination, occlusion etc. In this paper first, we apply Scale Invariant Feature Transform (SIFT) method on image to extract key points and the corresponding descriptors. Then we represent the extracted descriptors in form of AG graphs. So scene image and the model of objects we have two different graphs refer to it as scene and model graph. We suggest using the relaxation labeling to the graph of scene against that of model. The result of experiment shows that using contextual information improves the descriptor matching significantly.