The Application Of General Contour In Affine Invariant Feature Extraction

The extraction of image affine invariant feature has important significance in image recognization and parameter recovery. In real world, the object is often distorted for observing under various orientations. In computer processing, we use affine transformation instead of this deformation. The affine transformation reflects the invariants of the object under the deformation; it is used widely in computer vision and image processing.At present, the feature extraction approaches can be mainly classified into two main categories:region-based methods and contour-based methods. The region-based methods consider the information of the local region, reflect the local characters and they are favorable effect, but these methods usually take high computational demands. The contour-based methods is based on the contour of the object, it takes small calculation, but relies on the extraction of the contour.In this thesis, we introduce the extraction of the general contour and its properties. We extract the affine invariants through the general contour. This method ia a novel method by combining the contour-based methods with region-based methods. By central projection transformation, the object can be transformed from region into a closed contour named general contour. The general contour keeps affine features; it’s very robust and can be easily extracted.Main results of this thesis are as follows:(1) Propose a novel affine invariant descriptor based on general contour. We find the invariants by cutting the object into some slices which have affine characters through the general contour.20gray images have been tested in image recognization by our method. The result shows the effectiveness of the proposed method. By compare with AMIs and MSA, our method seems more robust to noise.(2) Improve the construction of the expand centroid by general contour. The new centroid can be used in binary images. Then we find more affine invariant points to solve the affine parameters. Some binary and gray images are tested in the image recover experiment. The result achieves better effect.