This software is based on a model-matching idea known as chamfer matching. The model is a
binary image such as that shown in Fig. 1a. In chamfer matching it is required to find the model in
an image, such as  Fig. 1b or 1c. The method first finds weak and noisy edges and removes them.
Examples showing edges of Figs. 1b and 1c are shown in Figs. 2a and 2b, respectively. Then, the
distance transform of the obtained edges is determined. A pixel value in a distance transform
image is proportional to the distance of that pixel to the edge pixel closest to it. Distance
transforms of Figs. 2a and 2b are shown in Figs. 3a and 3b, respectively. The model is then
shifted over the distance transform image and at each shift position, the sum of distances at the
model pixels is determined, and the shift position producing the smallest sum is chosen as the
best-match position of the model in the image. The best-match position of the model of Fig. 1a in
Fig. 1b is shown in Fig. 4a. Fig. 4b shows the best-match position of the model in Fig. 1c.

Using chamfer matching, regions may be selected in one image and searched for in another
image, thereby establishing correspondence between regions in the images. If closed regions are
used, the centroids of corresponding regions may be used as corresponding points to find the
transformation parameters and register the images.

Chamfer matching works well when the model and the image do not have rotation and scaling
differences.
(a)
(b)                                                                                              (c)
Fig. 1. (a) A binary image showing a model to be searched for in image (b) or (c).
(a)                                                                                               (b)
Fig. 2. (a) Color edges of the images in Figs. 1b. (b) Edges of the image in Fig. 1c.
(a)                                                                                                         (b)
Fig. 3. (a), (b) Euclidean distance transforms of the images in Figs. 2a and 2b, respectively.
(a)                                                                                              (b)
Fig. 4. (a), (b) Chamfer matching results when using the model in Fig. 1a and the images in Figs. 1b, and
1c. The best-match position of the model in each image is shown by overlaying the model with the image.
To obtain a license for this software, place and order here =>
2-D chamfer matching
Image Fusion Systems Research