The project contributes to the automation of a procedure for medical diagnosis in whole body photo imaging, which are used in the early detection of potentially deadly skin cancers. The objective of this module is the detection of correspondence between points located in two similar images (matching). This task is widely implemented in “Computer Vision”, for example in “Automatic Image Stitching”. The following link provides source code and executable application for “Automatic Image Stitching”.

http://www.codeproject.com/KB/recipes/automatic_panoramas.aspx

I applied the correlation module from the Accord.Net library to images of our project, and I got the result shown in next figure. This figure shows 29 matching. Among these matching we find good and bad matches.

Fortunately, in “Computer Vision” there is a widely used method for cleaning results as those in the last figure, this method is called RANSAC (Random Sample Consensus). We used RANSAC to extract from last figure the projective transformation between the two images. I applied this method and got the result of the next figure, where only 16 links have survived. The 16 points are called points of consensus. It is interesting to verify the 16 matches that were found. I think that only one of them is a mistake. I must combine these points with the another technique based in traces for a better automatic registration.