M. Pinzolas, R. Chumilla, J. López-Coronado (Spain)
Computer Vision, Object Recognition, Graph Matching
In this work, a new algorithm for object recognition is presented. This algorithm, called Affine-invariant Graph Matching (AGM), is able to recognize objects that appear translated, scaled or rotated in the image plane in non structured images. It is also very robust against illumination changes, and it can deal with slight deformations of the model and partial occlusions. All of this is performed without neither an extensive use of models nor exhaustive comparisons. For testing purposes, a set of pictures of several geometric objects taken from a stereo head against different backgrounds and illumination conditions has been used. In this set, objects are rotated and scaled with respect to the models. Also, the algorithm flexibility is studied in an experiment in which hand-sketched approximations of the objects are presented for recognition. Results show a high accuracy in object recognition, and in the correct determination of the scale and rotation angle of the objects with respect to the models.
Important Links:
Go Back
