Affine Invariant Shape Descriptor Using Object Area Normalization

Shape features play vital role in computer vision applications. The object silhouettes provide more evident properties for shape matching, and analysis. This paper presents contour and region based affine invariant shape descriptor based on object area normalization. The continuous contour is normalized by dividing total object area into equal part areas using sector area approach. This forms the 1-D feature vector to represent image object characteristics. The correlation coefficient and Euclidean distance are the metrics used for similarity measurement. The proposed method is validated on MPEG-7 CE Shape-1 Part-B dataset images and its affine variants. From the experimental results, it is observed that normalization based on area of the sector is more accurate in representing contour and region information than triangle area method. Moreover, this work can be adapted in object recognition and image retrieval tasks.