Shape retrieval based on manifold learning by fusion of dissimilarity measures

Content-based image retrieval (CBIR) is one of the most important research areas with applications in digital libraries, multimedia databases and the internet. Colour, texture, shape and spatial relations between objects are major features used in retrieval. Shape features are powerful clues for object identification. In this study, for improving retrieval accuracy, dissimilarities of contour and region-based shape retrieval methods were used. It is assumed that the fusion of two categories of shape description causes a considerable improvement in retrieval performance. The main goal in this study is to propose a new feature vector to coincide semantic and Euclidean distances. To accomplish this, the desired topological manifold was learnt by a distance-driven non-linear feature extraction method. The experiments showed that the geometrical distances between the samples on the manifold space are more related to their semantic distance. The proposed method was compared with other well-known approaches by MPEG-7 part B and Fish shape data sets. The results confirmed the effectiveness and validity of the proposed method.