Analogy with Object Recognition in Computer Vision

The problems raised in the research of the proteins 3D structure have a surprising similarity to problems in computer vision.
The task of model-based recognition in computer vision is given a model database to identify and locate in the target image (see figure [13.6]) all the instances of models. The models appearing in the image are usually transformed by an a-priori unknown transformation, they can also be partially occluded. The solution is shown in figure [13.7].

  

Figure: model and target images from [8]

  

Figure 13.7: The solution

One can divide the problem of structural comparison into two major tasks:

• The correspondence task - detecting matching features (difficult)
• The best superposition of given matching features - finding transformation of one structure into another with minimal RMSD (Root Mean Square Deviation)