Rotation invariant Shape Contexts based on 2D Fourier transform and Eigenshapes for Radiological image retrieval

This paper presents a new descriptor based on Shape Contexts, Fourier transform and Eigenshapes for radiological medical image retrieval. First shape context histograms are computed. Then, 2D FFT are performed on each 2D histogram to achieve rotation invariance. Finally, histograms are projected onto a more representative and lower dimensionality feature space that highlights the most important variations between shapes by computing eigenshapes. Eigenshapes are the more representative features, they are the principal components for radiological images. The proposed approach is translation, scale and rotation invariant, furthermore, retrieving operation is robust and fast due to space dimensionality. Experimental results with classes of the medical IRMA database demonstrate that the proposed approach produces better performance than known Rotation Invariant Shape Contexts based on Feature-space Fourier transformation. keywords: Image retrieval; Shape Contexts; Fourier transform; Eigenshapes; Radiological images.