# Signal processing vs image processing?

Although both are domains of EE, but what are differences between them? I am able to draw following conclusions from wikipedia article of dip and also from some other web reources

1) DIP(Digital image processing) is a subset of DSP(digital signal processing)

2)Because of above concept in (1), DIP only deals with 2 dimensional images/signals or in some case 3 dimensional images/signals(forexample RGB images) while in DSP there isn't any restriction on number of dimensions?

Please kindly guide me and correct me if i am wrong in my concepts/assumptions

• No, No, and No. Both topics are incredibly broad. And yes there is some overlap between them. But there are things in DIP that are not in DSP and vice versa. Perhaps if you posted the course descriptions for the two course then folks could give you better answers. – scorpdaddy May 1 '20 at 21:19
• dsp.stackexchange.com – Rodrigo de Azevedo May 2 '20 at 5:21

Images are signals, and like any other signal can be processed digitally or in continuous-time. So, yes (1) is correct.

How images differ from, say, an audio stream is that discrete points on an image have a 2-dimensional (x and y) spatial pattern (pixels) that comes into consideration when processing them, while audio is considered to be 1-D, with samples only separated by time.

If a signal is considered to be a vector it has more than one dimension (in a vector signal analyzer, magnitude and phase) and so the DSP algorithm is expanded to deal with that.

So the first part of (2) is correct. The second part about ‘3-D’ is not.

RGB isn’t ‘3-D’. It’s just more bits that define a pixel sample. 3-D with respect to images can mean stereo imaging, or it can mean temporal processing over a range of images to, for example, detect motion. So it’s important to clarify what you mean when you say ‘3-D image Processing’.

So you could have 3-D stereo images that are processed in sequence to detect motion in 3-D space... so 4 dimensions in that case.

Conclusion 1 is correct.

Conclusion 2 doesn't seem wrong, but I don't find it very useful.

Unless I'm mistaken, what you are really want to ask is what is the difference in content between the courses.

If the DSP course is not a pre-requisite of the DIP course then the difference is probably that image processing uses a bunch of techniques that aren't commonly encountered in other applications due to their multi-dimensional nature.

More general signal processing is more focused on encoding and decoding data from one-dimensional waveforms (or multiple one-dimensional waveforms simultaneously).

True, DSP isn't technically limited to the number of dimensions but in a DSP course I would expect you to focus a lot more on one-dimensional signals than you would in DIP...at least until you get much more advanced where you're looking at multi-dimensional DSP in general rather than just focusing on image processing.

It's probably like the difference between a science class and a physics, chemistry, or biology class of the same level. One is more focused and in depth while the other covers a broader range of material.

Disclaimer: I don't specialize in signal processing, or even work in it.

• I really want to ask is what is the difference between the both fields DIP & DSP , not the difference in content between the courses – DIP_fan May 2 '20 at 17:31

Having assisted in developing an Autonomous Vehicle decades ago [ back when UMASS_Amherst predicted "the vision problem will be solved when TeraOp/Sec machines are available"], and pondered on the information-flows (information input from imagers, and the information output to steering) in such machines, I'll offer some thoughts.

The computational load for successful detailed 3_D road traveling is far in excess of what our silicon imagers provide (the retina has 7 layers of Edge Detection, etal along with rods and cones). I've chatted with various present-day AV people, they commenting on how the mobile machines freeze up in intensely-busy scenes such as turning in an intersection with blowing paper.

Thus the vast uncertainly in modern imagers and the associated compute power is a problem, resolved by identifying the regions of uncertainty and revisiting/foviating on those regions, perhaps at a faster frame rate.

The advantage of the human imager is our in-situ and continuous Edge Detection and Texture Identification.

Summary: image processing may need { we have LOSS_OF_INFORMATION_HERE , thus vast risks ) recovery behaviors, such behaviors not needed in classic 1_D problems.