The primary reason for shape characterization and matching is to use it for characterization and recognition of the associated objects. However, the shapes obtained from segmentation and/or edge detection of real world images are, at best, approximations of the actual shapes of objects. Unsupervised segmentations often deviate from object boundaries to include parts of other objects or background. Similarly, objects of interest may be partially occluded by other objects in natural scenes. We address the problem of adapting known shape characterization and retrieval methods to make them robust to errors in the basic input – the binarised shape image corresponding to an object. An effort is made to retain the ability to deal with scale, rotation and translation. The presented method is based on the centroid distance shape signature, but which does not sample the perimeter points evenly along the perimeter length. Instead, the sampling is done evenly using an angular measure. This property of our signature localizes the changes due to occlusion. For similar reasons, we do not derive a shape descriptor where each feature potentially depends on the entire shape signature. The onus of achieving various invariances is shifted to the definition of our similarity metric. Again, to take care of the changes in the perimeter, the similarity measure has been designed to produce small changes for small segmentation errors. The approach presented here can be applied to many applications such as Content Based Image Retrieval (CBIR), Target Detection, Medical Imaging etc. The limitations of the method are its inability to deal with complex shapes that have perforations, tendrils etc. Index terms: Shape, Shape signature, Similarity measure, Occlusion.