3D model comparison is an issue that has received wide attention from researchers in computer graphics, vision, and database systems. Most of the previous approaches have been based on a common strategy that characterizes a 3D shape as a high-dimensional feature vector, and measures the similarity of shapes in terms of the distance between points in the high-dimensional feature space. These methods often suffer from erroneous cases where feature vectors do not properly discriminate between models belonging to different categories (eg., animals and plants).
This paper presents an alternative approach to 3D model comparison. Our method compares the multiview images representing the appearances of 3D models, and computes the similarity of the models based on the similarity of the images. In comparing a pair of images, a morphing function is constructed, which describes a continuous transformation from one image to another. This image-based approach enables us to group together morphologically similar models while distinguishing between apparently dissimilar models. Experimental results show that our method is suitable for comparing soft objects such as animals.