Shape retrieval using combined Fourier features

Centroid-distance signature is one of the classical shape signatures which are used to derive Fourier descriptors. It has the strong ability of capturing the global features of the shape. However, its ability of characterizing the local shape features is poor. To overcome this limitation, a novel shape signature, perimeter area function (PAF), is proposed in this paper. PAF places all the vertexes of the triangle on the shape boundary. Therefore it can finely capture the local shape boundary information. The Fourier features derived from the proposed PAF are combined with the ones derived from the traditional centroid distance signature for shape retrieval. This hybrid scheme presents an accurate shape description which is more suitable for shape retrieval than the traditional Fourier descriptors. This method is tested on the standard MPEG-7 shape database and the proposed leaf shape database using the standard performance evaluation scheme. All the experimental results indicate that the proposed method outperforms the existing Fourier descriptors and Wavelet Fourier descriptor.

Research Highlights
► Perimeter-area function (PAF) can finely reflect the local features of the shape.
► PAF can capture the convex and concave properties of the shape boundary.
► The combined Fourier features is generated from PAF and centroid distance (CD).
► The combined Fourier features outperform the existing Fourier descriptors.