Image Registration

In the first week, we will describe and implement geometric transformations of images. We will learn how the transformations and their compositions work. We will also implement the translation estimation using phase correlation in the frequency domain.

Download the support package for the 1st week: registration1.zip

After we will discuss the geometric transformations you will try to transform images using translation, scale and rotation.

Using the provided example, try to generate image from A.png which is first scaled then translated (the result should look like A_st.png). Do it two ways: 1) scale an image using scaling transformation, then translate it using translation. 2) compute a composition of transforms and create an image in one step.

Finally, the main function to implement in the 1st week will be phase_corr.m, the function which estimates translation between two images.

In the second week, we will continue with estimating also rotation and scale. Download the support package for the 2nd week: registration2.zip. See the example of result here.

When you are done, upload the ZIP-archive containing the following m-files to brute:

phase_corr.m,
log_polar.m,
fourier_mellin.m

The Brute upload system will generate visualization of results for your implementation. You may compare your results to ours.