Scene matching using F-histogram-based features with possibilistic C-means optimization
Fuzzy Sets and Systems
| Hi-index | 0.00 |
Affine invariant descriptors have been widely used for recognizing objects regardless of their position, size and orientation in space. Examples of color, texture, and shape descriptors abound in the literature. However, many tasks in computer vision require the consideration of the objects' spatial relationships. Earlier work showed that the relative position of two objects could be quantitatively described by histograms of forces. Consider images of a scene captured from different viewing geometries. The images contain a number of objects, and the spatial relationships among these objects are captured using force histograms. These force histograms are used as the image spatial relations descriptors. In previous work, we studied how affine transformations affect the force histograms descriptors and how to approximate the parameters of an affine transformation that best relates two descriptors. We now propose a method for generating the correspondence map between two image descriptors. The confidence of each map is calculated using features calculated from the estimated affine transformation parameters. From this map, we generate the object correspondence confidence matrix that contains the degree of confidence for relating objects in the first image to objects in the second. From this matrix, we generate a one-to-one object correspondence map for the two images by maximizing the aggregate object correspondence confidence. We compare this new approach to an existing method based on the Fourrier-Mellin and log-polar transform. We demonstrate possible applications of this approach in object registration, scene change detection, and matching hand-drawn sketches to real images. We perturb the image by adding, removing, displacing, scaling, and translating individual objects in the image to evaluate the robustness of our approach.