engleski

A morphological signature transform for shape description

A novel shape description method based on the multiresolution morphological image processing is presented in this paper. In this method a representation based on the areas of the input binary image successively eroded by multiple rotated structuring elements at different resolutions is used. The proposed method has the translation, the rotation, and the size invariance property. The binary image which contains the object shape to be described is represented by means of multiresolution pyramid. The primary structuring element is selected and used to generate a set of rotated versions of the primary structuring element. The successive morphological erosions of the input image at different image resolutions by multiple rotated versions of the primary structuring element are obtained. The successively eroded images are called the signatures and are used to describe the original object. The process of obtaining the signatures is called the Morphological Signature Transform (MST). The signature areas are computed for each rotated structuring element and each input image multiresolution pyramid level. The areas are normalized across the pyramid levels to enable more accurate matching (distance measure) between object descriptors. The obtained set of numbers for each of the rotated structuring elements is formed into a vector. The vectors are circularly shifted to achieve rotational invariance of the shape description method. The vectors obtained in such a way are then used as the shape descriptors. One of the basic underlying ideas of this work is the 'divide and conquer' principle applied to shape properties. Shape description methods often attempt to directly extract shape properties from a complex shape. The drawback of such approaches is that the extraction of complex shape properties may not be efficient. The approach taken in this work is to transform the initial single complex object to a set of multiple objects having simple shape properties. This process can be viewed as an extraction of different shape properties or shape property decomposition. Shape properties from multiple transformed images (signatures) are extracted in order to obtain shape descriptors. The advantage of such an approach is that it extracts simple shape properties from complex objects. However, several transformed images will enable simple shape descriptors to accurately describe the given initial shape. Experiments have been performed to investigate noise robustness, invariance to rotation and scale distortion of the proposed method, and have shown that the method is robust to noise and invariant to both translation, rotation, and scale change. A set of test images has been created, corrupted by noise, scaled, and rotated. The Euclidean distance between the shape descriptors for the original and the corrupted images has been computed, and has shown that the proposed method has good shape matching properties. The proposed shape representation method is applied to the problem of shape matching which evolves in many object recognition applications. Here, an unknown object must be matched to a set of known objects in order to classify it into one of finite number of possible classes. In real world applications, the unknown object may be corrupted by noise. The results for images corrupted by noise, scale, and rotation deformation are presented and discussed.

shape description; mathematical morphology; image analysis

nije evidentirano