Objects matching combining color and shape

This paper presents a general 2D object characterization and matching scheme based on the information provided by color and shape. We focus on matching objects from generic images with complex scenes. In order to identify the region associated to each object, we use an unsupervised segmentation process based on a hierarchical representation of the image. The image is characterized using a uniform-scale chromatic space for the color, and morphological filters and deformable models for the shape and outline, respectively. Our method has been tested in order to retrieve objects of various colors and shapes. The proposed scheme is invariant to image rotation, affine transformations changes in illumination, and robust to small perspective and points of view deformations.