Image Processing is best choice to do Master Thesis because many options are available in Image Processing like Surveillance, Medical Imaging, Satellite Images etc.
Instruction level parallelism is the good topic for research at master level. A lot of research is being done in this field. The researchers are continuously trying to exploit ILP by using aggressive techniques. ILP is the technique by which a sequential program is executing in parallel without re-writing the program (using parallel algorithm).
Agreed with @Rajendra Kumar; in addition, I would like to suggest to take a look at the compiler optimization techniques, while it is totally in the scope of the Computer Arch - and High performance domain, it will give an insight of the vast usage of machine learning techniques, working on embedded devices and parallel-prog as well. The overall problem is NP-hard as it is a mixture of scheduling and assignment, so lots of gaps are still there in the literature,.
I also agreed with @Amir Hossein, in compiler optimization you can cover the milestones in computer architecture and that will open a great industrial/research carrier.
All three of those areas have interesting problems to solve and job prospects. Which do you find most interesting? Think of it this way - you will be working on whatever problem you choose for at least the next year or two and, if all goes well, related problems for much longer.
The first step when selecting a thesis problem is to speak to your advisor. I'm not familiar with the process at your university, but if you do not already have a thesis advisor then find one as soon as possible. Speak to professors who do research in these areas and ask them what problems need to be solved and are a good fit for a master's thesis. Speak to a professor after a class in one of these fields. Researchers are always looking for students with initiative. You need a good fit between yourself, your advisor, and your thesis topic to be truly successful.
Image Processing is best choice to do Master Thesis because many options are available in Image Processing like Surveillance, Medical Imaging, Satellite Images etc.