How Image Processing and machine learning can be linked together?

Machine Learning basically means that you're training the machine to do something(here, image processing) by providing set of training data's. Machine Learning have models/architectures, loss functions and several approaches that can be used to determine which would provide better image processing. Also, it depends on the type of image processing you intend to do as there are certain loss functions that perform better than other due to their inherent properties for example there's high possibility that cross-entropy loss function could perform better than other loss function to give a better image processing. Hopefully, this helps.

Also, I believe that CNN architectures would perform better.

It depends on what you want to do. There are so many examples to build your data and learn them to find the best solution for your work.

Deep Learning for Medical Image Processing: Overview, Challenges and Future

Machine learning used for improving image processing results for exampl. By feature extraction and classifieled.

In general the (pre-) processing of an image is often an initial step to later extract the features that would be used to train a machine learning classifier. Signal processing can be used to enhance or eliminate properties of the image that could improve the performance of the machine learning algorithm.

Image processing problem => Optimisation problem.

Machine learning => Effective tool to solve Optimisation problem.

So, they linked together.

How? You can read this overview presentation http://www.ee.pdx.edu/~mperkows/CLASS_ROBOTICS/FEBR26-2004/ROBOT-DECISION-TREE/MLforIP.ppt

Whenever there is a image recognition/classification problem, Machine learning is there to solve it.

e.g. Face identification, Face recognition, Facial expression recognition, Tumor/disease detection from medical images, Car licence plate recognition, optical character recognition, and so on