|
  
Next:Combining results from searching Up:Model Validation Previous:Feature Search on the
Raw Pixel Search Search at the raw pixel level consists of comparing the pixels of a rectangular template of size  with each sub-image of the same size. The process is called template matching. We use the correlation
coefficient as an indicator of similarity between the template and the sub-images. The correlation coefficient is defined as the cross-correlation between properly normalized versions of the sub-image  and of the template T, of size  : 
where  .
The correlation coefficient is translation invariant by definition. However, it is not rotational and scale invariant. As a result, the same template needs to be correlated with the same images at different scale
and different rotations in order to overcome this difficulty. Furthermore, template matching is rarely exact as a result of image noise, quantization effects and differences between images of the same region.
For example, seasonal changes introduce effects that make it difficult to match satellite images. |
