An efficient contour based shape descriptor for representation and classification of medical images

Abstract EN

Foundation of a robust image classifier depends on having an accurate image’s description based on its visual content.

Recently, contour-based shape descriptors are employed for object-based representation due to its applicability to generic shapes and tolerance to noise and distortion.

In this paper, an approach for extracting a hierarchical contour-based shape descriptor is described.

It benefits compactness, flexibility, and suitability for image classification.

This approach mingles the global description of moments with the multiresolution decomposition property of wavelet transform for robust shape representation and extracts approximate information about object’s shape.

Furthermore, by integrating a suitable segmentation algorithm into the descriptor generation schema, the proposed approach turns into object-based.

Experimental results show that the resultant descriptor provides promising results in terms of accuracy for medical image classification.