← Previous lessonFree lesson 2 of 6 · lesson 2 of 48Next lesson →

Unlock all 48 lessons of Advanced Medical Image Processing in Clinical ApplicationsFull course ₹40,000 · paid

Price & syllabus
Free lesson · 34 min · no payment

Stages of Image Processing, System Components and Image Formation

A free lesson from Skill-Lync's Advanced Medical Image Processing in Clinical Applications course, on Grow by Skill-Lync · Last updated 16 Nov 2026 · Session: “Module 1 Session 2”

What will you learn in this free Advanced Medical Image Processing in Clinical Applications lesson?

The lesson walks through the image processing pipeline from acquisition and filtering to enhancement and restoration. It covers system components such as GPUs, frame grabbers and software like ImageJ and 3D Slicer, then image formation as illumination times reflectance.

  • Pipeline: acquisition, filtering, enhancement and restoration
  • Hardware: frame grabbers, GPUs and cloud notebooks
  • Software such as ImageJ and 3D Slicer
  • Illumination and reflectance in forming an image

What comes after lesson 2?

The next free lesson is lesson 3, “Sampling, Quantization and Reading Images in Python” (22 min). Later in Skill-Lync's Advanced Medical Image Processing in Clinical Applications course (paid): lesson 4, Module 1 Session 4; lesson 5, Module 2 Session 1; lesson 6, Module 2 Session 2. The full course has 48 lessons (21 h 26 min of video); see the course overview or its price and syllabus.

Lesson transcript

Machine transcript of the lesson audio (lightly cleaned). Technical terms may contain errors; the video is the reference.

Hello and welcome back to the session 2 of week 1. As a brief recap to what we have seen in session 1, we have seen what digital image processing is, what a digital image is, and how one goes about in quantizing and sampling an image, an analog image, to get a representation which can be used in a computer. Now once we understand how an image is acquired, and also various kinds of frequency ranges in which the electromagnetic wave falls and is being captured, we now go into the next step of what are the various stages in which an image gets processed.

So for this we have a very nice block diagram to explain the interaction of the various steps. This image can be understood by looking at it in a cyclic order, starting from the left and moving top up and then right. So initially we can take a problem domain. The problem domain could be medical imaging. We have maybe a CT, maybe an X-ray, maybe an ultrasound image. That image should be first acquired. And we saw also examples in the previous session about direct image versus indirect image.

And many of the medical images are indirect image because you need a reconstruction stage for the image to be formed. So that comes under the image acquisition part of the image. Now once the image is acquired, the next step is to go about filtering. If there are any noise, you want the noise to be removed. And then there is aspect about enhancement. So you then make the image look better, sharper, brighter, all those come under enhancement. There are aspects about image restoration which comes in some scenarios and these come when you have issues due to say blurring.

You have a fast moving object and you are trying to image it and that comes as blurry. Then you need to do restoration process to correct the blur. So there are many different restoration aspects to correct any acquisition related problems. There is also a whole domain of color image processing which has now expanded into a field in itself. Color is a very important attribute in an image which can be leveraged to get much insights into the image. And it initially just started with RGB, but there are very many color models like LAB, HSV and so on.

The next step in this are wavelets and transforms. Here we don't analyze the image in the spatial domain, but we convert the image into another domain called the frequency domain and analyze the image in the frequency domain. There are advantages in operating on the frequency domain and we would see some of the advantages when we get into looking into details of the frequency image processing. Compression and watermarking are very important attributes to reduce the extent of information to be processed. So when you are sending an image over say internet or via the recent modalities like WhatsApp, you don't want to send a high resolution image with gigabytes of data.

So then compression is actively employed to make the image size smaller without losing much detail. And these are inherently done even when you send an image, for example, by WhatsApp from you to somebody else. The image is inherently compressed to save on bandwidth. Watermarking is another important aspect which comes under copyright issues. How do you say a given image is belonging to you? And for that, watermarking is an approach where you kind of put a digital footprint on the image, sometimes visible, sometimes invisible, to mention copyright that it belongs to you or a particular organization.

These steps that we have covered so far, starting from acquisition till compression, watermarking, have input as image and output as images. And we have already known that this broadly comes under image processing or basic level image processing as we saw. Now going to intermediate and advanced level image processing is when input are images and output we see are attributes. As I mentioned in the previous slides, attributes could be number of students sitting in a class or number of cars in an automated driving scenario, operate on images but produce outputs as attributes.

This also has an overlapping aspect with computer vision. Many of these advanced aspects in segmentation feature extraction goes into computer vision aspects also. So we start with morphological processing, which are basically set theory operations and progress into segmentation, which is breaking an image into integral components. What are the various objects, structures present in the image? We also look at feature extraction. What are the key things, important elements in an image, right? Those important elements are called features. They can be corners, they can be corners, they can be edges, they can be texture, and so on.

Finally, we have pattern classification. You want to classify a given texture to be belonging to an ocean or maybe a given object to be belonging to a particular class. So it can also arise with object classification and so on. So I hope with this broad outlay of the various steps in image processing is clear. And any image processing course goes into some level of detail in each of these aspects. So the explanation for each of these steps are already done while covering that graph. But to make it a little more detailed, here we see acquisition, it's about having a sensor to get an image.

It can be a sensor like a mobile phone camera or your point and shoot camera or a SLR camera for the computer vision image processing kind of thing. And for medical imaging, it is more about ultrasound system, x-ray systems or CT systems, which are sensors to acquire the image. I mentioned about enhancement, which is about making the image more sharper and brighter. And we look at mechanism like contrast enhancement, histogram equalization and making the image more brighter, sharper. The restoration part I already mentioned, an image could have blurring cost due to fast moving object, could be uneven illumination because light is not uniformly thrown on an object or reflected from an object.

And so rectifying these distortions is broadly falling under the image restoration part. We also saw about transforms. Here we mentioned wavelets because that is one of the predominant main frequency domain processing. And wavelets help in representing the image in various degrees of resolution. And what do we talk about resolution is what is the frequency content present in a given image, right? And that information one can extract by using wavelets. I also briefly alluded to compression and that is to reduce the size of an image for communicating it over some channels like internet, like an email or a WhatsApp.

And there are possible loss which occurs due to compression. And we can also employ techniques which don't lead to loss. And hence we have lossy and lossless compression. Morphological processing is also a huge domain which has kind of developed into a domain speciality in itself. It is based on set theory operations. The basic operations are very simple, but what one can achieve with morphology operations is actually mind-blowing. The context being that one can do very quick, very fast processing with morphological processing as compared to traditional methods like applying masking, mask operations or frequency domain operations.

So some of the methods are erosion and dilation, which we would see when we cover in detail about morphological operation. Segmentation, I had mentioned, is looking at various structures present in image, partitioning image into its various parts. So for example, if you have a classroom image, segmenting each of the or identifying each of the participants, maybe students could fall under segmentation. Feature extraction, as I mentioned, is about finding key elements in an image which makes it a particular category. For example, you are looking at a dog.

What makes this dog image is could be because of its nose, its eyes, the shape of the face, the ears, the tail, and each of these are features. That's a feature for an object. More low-level features are something like corners, edges, and maybe structures like circles, ellipses that you find. Image pattern classification is even more complex. You kind of look at objects and its descriptors. So as I mentioned, if I give you an image, how do you say there is a cat in the image? So you need to object localization to know where a cat-like structure is present.

And then you also... Need classification to know how do you classify it as particular objects like say a cat. So here you use correlation methods, neural networks, and this kind of also has overlapping regions with the machine learning and you can say artificial intelligence. So now that we have understood the various steps of image processing, we now look into image processing systems. A system is nothing but as components which help in getting a task acquired, right? An image processing, digital image processing doesn't only use a computer and an image store on the hard disk to process, right?

There's a lot of steps between acquisition, storage, transmission, which also come as part of the image processing system, right? So we now look at the overall system and understand what else other than a simple image and a computer generally place in an image processing system. And I always give examples for relating it to medical image. So to make it more specific to our course requirement of medical image processing, right? So our problem domain here again is maybe acquiring a X-ray and trying to understand what that X-ray is communicating or showing as a disease and processing that X-ray to maybe send it to another specialist to get his opinion, right?

So here we have sensors. In this case, X-ray, it can be an X-ray collimator or an X-ray film and so on, which are the sensors where A, a sensor which emits the X-ray light, B, which can receive the X-ray and form the image, right? So the image sensor forms the image, but you also need hardware to construct. Like as I said, in indirect image processing, you need some post-processing or reconstruction to create the image. And these recons are generally algorithms which are based on some research which have happened or you used specialized softwares which have the recon capabilities to do the mapping of the sensor data to form an image, right?

The left you see more hard copy. So if you want to create multiple copies of it, either print copies or soft copies, you also need hardware and mechanism to create copies of that image. And then once you have the image ready with the hardware, with the sensors, now comes the computer where you can visualize it. So the computer stores it. It may store it in the hard disk. It may store it somewhere else in the cloud. And then you also need to see what you have acquired, maybe a display like a monitor or AR-VR.

These days, there is a lot of interest and research happening in augmented reality, virtual reality, where you get an image and it is directly showcased on a head mount display. And finally, you may need to communicate this X-ray image to a specialist to say, I suspect there is pneumonia in this patient. What do you think? So for this, we need a network to communicate this image to another tertiary hospital and then get feedback from them to say, yes, what you're suspecting is correct or no, it's not anything anomaly.

It is all fine. The patient can go home. So if you see an image processing system is a conglomeration of many things, not just an image sitting on a computer. And it has multiple steps from defining where the problem domain is to what sensors are being used, what softwares are being used, what storages we use, how do we display it and so on. And depending on the problem you work on and the challenge you work, there might be challenges even in the sensor, even in the software.

It could be in the displays, it could be on the cloud, the transmission. So there are various challenges which arise and some aspects of it we cover in our solving a digital image processing on a computer. But the aspect of this slide is that image processing systems have much larger coverage in the whole process of image processing. So whatever I explained in the last slide, we now go in detail in this slide with putting some more details about the various components. And as I mentioned, the sensors convert whatever light energy, if you remember our introduction, what we said, what is an image?

An image is a multidimensional signal and the signal is generally an electromagnetic spectrum, right? Electromagnetic wave. Generally, this is the case in the visible image. And so you have a multidimensional electromagnetic signal and that needs to be converted into a digital format. We saw about sampling and quantization and these aspects about sampling and quantizations are done by digitizers. So the digitizer could be sitting right in your digital camera, could be a digitizer could be indirectly used when recons are being done on a medical image part.

We also then came to the processing hardware. If you come to challenging scenarios like video, video images are captured at very high frame rates, right? You process maybe 30 frames in one second, then you can't do such high-end processing on a direct computer, but you need something called a frame grabber. You need to grab the images and stream it image by image to a computer so that it can process it, right? And coming to the recent trends, when you have very intensive computing for deep learning per se, there are specialized hardware called graphical processing units, which can do millions of operations on the graphic card directly.

These are specialized hardwares which can help aid computation, especially matrix computations in rapid manner. And when you go to deep learning and learn more about deep learning for computer vision and also deep learning for medical image, you would see a lot more usage of GPUs, which are now available to be used directly on the cloud. Like you have Google Collaboratory, which is a GPU provided by Google that you can directly utilize through a browser. The computer also is an integral part. We do realize that when you're doing image processing, you need to have a computer to process the image.

And it can be a normal laptop or a desktop. One can also imagine having a supercomputer if you're working on astronomical images like images of the galaxies or you're looking at microscopy images like looking at slide images, slide smear images in digital pathology. You would need really very high processing power, especially from a supercomputer. Well, computer doesn't suffice everything. You also need software to help you work on those images. There are very many softwares these days to operate on images to get the rudimentary tasks directly done without writing, needing to write code.

And these softwares range from Photoshop to there is something called ImageJ for medical images. And there's also Slicer 3D, which we'll see later in this presentation, which helps to go through those images. And Slicer also has an option to write code, Python code within it for if you find the suitability for what you want to do, not present in direct blocks in Slicer. You can write your own modules and integrate it into Slicer. Mass storage is about how do you store whatever you acquire. And in fact, storage has immensely improved in the last decade.

Now storage is available at very easy prices or very low prices, but still equivalently, the amount of data we generate from scanners have also exploded. Right. You now generate gigabytes of data from a CTE or an MR scanner. And if you are doing live imaging like real-time imaging, you can easily cross terabytes of data for a large region being scanned and high resolution being scanned. So then storage is also a very important aspect. And since medical hardware throw about gigabytes of data, we also need to archive them, which you are not using in maybe solid state drives or put them into the cloud.

Displays are an integral component also in image processing systems because finally an image is consumed by an end user like maybe a clinician by looking at display like a monitor or on a projector. And the importance of it is very, very key in how the image is being displayed because if it is improperly displayed, you can miss spotting a pathology. So the displays play a very important role. You cannot afford to have specks and dots or noise on the screen or on the wall which is being projected in case of a projection.

And that could be misled as being a pathology case. Right. Hard copy devices are more for creating copies of the image. You can use printers to print. You can write it in a CD and so on. And one could also print it on a film and then share it across. And last but not the least, to transmit the image to other sites, you can also use cloud to transfer the image from one location to the other. So let's step into the first aspect about image sensing.

So here we want to convert an analog signal into a digital signal. And as many. As I mentioned earlier, the digital sensors play a key role. There are various kinds of sensors. I mean, in a camera, you have earlier CMOS, and then you come now with CCD, a charge couple diode. So these are now sensing technologies which can help effectively, efficiently convert an analog into a digital image. So you see here in this image, what you see here, there is a sensing material which is represented in dark blue, which needs some power, because only when power is there, it can convert an incoming signal into a digital format, right?

There is energy coming from, say, electromagnetic spectrum or light falling on the sensor, and this sensor is sensitive to light. So the particles here, like the CCD, CMOS, react variably based on the amount of energy present on it. And then you use digitizing, digitizer to convert that sensed signals into a digital format, which we saw could be represented in a matrix form. So typically, there are two kinds of sensors. One is a one-dimensional sensor, which you go line by line, and you acquire one line of an image.

And then there is also an array sensor, which you acquire one short 2D matrix of information. So now we see first, how do we go about in a linear fashion, how do we compute an image, right? So an object is to be imaged, and for ease of understanding, we have a concept of a drum, which we would see in image for man in the subsequent slide. So light source for highlighting whatever structure is inside the drum, right? You have a cylinder case, and the light is inside it.

And then negative is placed on top of the drum. And light passes through the film, and whatever regions the light is allowed to pass, you would have that formed on your film, basically. Before, I mean, so the film modifies the light, whatever is passing through it, and then you have sensors to receive what is the final image which is getting formed. So to more visualize how is it, here is the visual way to look at a creation of a 2D image. And here we do it line by line, you could also enhance this to make it as matrix sensors, but the beginning of it was more linear sensing.

So you have a light source inside the drum, you have film put on top of it, and this is probably, you can think of it as a negative of an X-ray, right, which is placed on top of this drum. And then light is switched on, light passes through the film, some regions of the film allow the light to pass through, some don't allow, wherever it is allowed, it is captured by the sensor, and the image is then formed, right? So the image creation is an integral part for image formation, but most image processing work assume that this step is correctly done in the best possible way and move on.

In scenarios where you have control, then you can kind of limit the issues like noise or blur or non-uniform illumination right at the stage where the image is being formed. Now we come to the creation of a 3D image, right? A 3D image is slightly more complex because you have structures in 3D to be represented. This has been a major advancement in medical imaging especially because inherently anatomy structures, like if you are looking at the brain, if you are looking at the lung, if you are looking at the heart, they are all 3D structures.

And you cannot view slice by slice and make conclusions about what the 3D structure is. So to acquire a 3D, there is usually a process of reconstruction and this comes predominantly under indirect imaging, right? You capture some data, which is coded data, and you do a reconstruction process to convert that coded data into a series of images. This is called a stack of images. What you see in blue here is a stack of images. It's like a loaf of bread with many slices. Each slice captures a different portion of the 3D image.

So imagine you are looking at the brain and the brain can be represented as a cuboid. You are looking at the first slice of the brain, the second slice of the brain, 10th slide of the brain, and then maybe 50th slide of the brain. So each slice represents one portion, one axial portion, if you can call, of the 3D structure. Now what is done to do this coded representation of 3D is that here you see a cylindrical structure, but this could as well be a patient, right?

What you see here is a patient lying on a table and then sent inside a bore. The bore is this hollow that you find in a scanner. Now the scanner could be a CT scanner, could be an MR scanner, could be a PET or a SPECT scanner per se. So each have slightly different physics behind acquisition, and people generally spend a whole research career learning one modality, right? You spend four years learning the integrities of CT. You can spend four years doing research and PhD on MRI, and so similarly so for ultrasound.

Now one cannot gather all the details in an introductory course on medical imaging. We would touch upon briefly what are the various kinds of medical image formats, but more importantly, in 3D, one of the key things is that you have, especially in CT and MR, you have a source which is rotating, right? This is maybe like an X-ray source you can imagine, and the X-ray throws X-ray onto the object. Some of it is obstructed, the remaining is passed through, and then it falls on a film, right?

X-ray film or detector, right? So the detector then is able to know what the object is being imaged in that small arc of that region, and then the source keeps moving to the next possible location to image, right? It turns, that's why it is in a circular structure, and once you move a little bit, you again send another beam of X-ray, and the detector also is moved. They are diametrically opposite to each other, and then you get a next arc information of that object. Then again you move.

So you do this circular fashion in multiple ways of looking at it, or multiple angles of looking at it, and then there is a recon process to put this together in a way that can be viewed more in a classical way looking at slice by slice. You don't want to look at it in arc by arc, but you want to look at it in a slice by slice manner. This recon varies from CT, which they do radon transform, to MR, where they do Fourier-based, and so on.

So let's now move to the next section on digital image acquisition, and probably we can go a little more detail into how digital images are being acquired, right? We have already seen sensors are integral part. We have seen how 2D images are formed, and 3D images are formed. So here we have a simple example of a 2D image formation, right? The image could be understood by looking at the process that you see here, starting from the left and moving all the way in a circular fashion till the down, right?

So you have a charge coupled diode, which is called CCD in short, which are sensitive for light forming, right? We already saw in the past that there is a light source, and you place a film on top of it, the light passes through it, and then there is sensors, which are active to find out what are the regions which allow light to pass through, and those which don't allow light to pass through. Once the CCDs acquire the data, you kind of do analog to digital conversion, and that makes it digital, right?

This then forms the sampling quantization that we saw earlier. You also do some amount of amplification of the signals to increase the signal characteristics, and then it goes into the digital image processing pipeline, right? You then do lookup table conversions. These are more ways to represent the image. In lookup tables are more for displays. How can you show an image which has very wide range? You can capture an image with a very large dynamic range, right? What I mean by dynamic range is your quantization could be very, very big, right?

Not a coarse, but a very fine quantization. And displays may not be able to handle that level of granularity. Then you need to convert it in some way, and these are done using lookup tables. A range will come under a particular signal space in your display so that it can be shown. And then once you have that, you can then process the image. You could probably convert it into a DICOM. There are two kinds of DICOM. A DICOM, as I mentioned in the introduction, is a medical image format representation.

We can create a secondary DICOM where you acquire an image and convert it into DICOM. A primary DICOM is when you acquire an image directly by. And it converts it into DICOM as it is being acquired. More details of it you would know when you get into details of DICOM. And then what is PAX is a storage in a hospital, picture archiving and communication system, which is more like a storage to store the image on a hard drive or to communicate it via a network to another hospital.

So we have seen the overall image formation model. And this model helps in converting our electromagnetic spectrum, which is a two-dimensional or a multi-dimensional signal, which we saw earlier as F of XY into matrix format, which is then digitized, quantized. And also, you have now done two kinds, sampling and quantization, digitized on the XY coordinates, digitized on your amplitude region, and then you have a digital image that you have with you. And since an energy cannot be negative, your pixel values will always be greater than zero, and it will always have finite amount of energy or a finite value.

Hence, your amplitude values are always between zero and infinity. So you have two components in an image. One is amount of illumination, which is incident, and the second aspect is the reflectance, right? So here you see illumination and reflectance. So what are the two things? How do you understand? So assume you have a chair in a room. To see the chair in the room, you need two things. One, you need light. The light should be either coming from a bulb or you have a torch. Let's assume it's a dark room and you're shining a torch on the chair.

So the torch you do switch on and shine on the chair is the illumination part. You illuminate the chair. But assuming that the chair is a structure which absorbs all the light, then you still won't see the chair. So the chair takes in some amount of light and then reflects back some amount of the light. Whatever is getting reflected falls on your eye and you then see that as, okay, here is a chair. The light could be, so if you see a red chair, that is because all other colors are being absorbed by the structure or texture present on the chair and only red is being emitted back.

When you throw white light, it doesn't mean your object will also show as white. It depends on whatever color the object is. It absorbs all other colors and reflects back only one of the colors. And hence you see a red chair because it reflects back the red light. So an image in looking at in a different way is a combination of two different signals which interplay illumination from your torch or your bulb and reflectance of the object when it shines back, right? So these two combine together to form an image and hence an image can also be represented as a multiplication of two components.

Illumination at each location, how much light is being shined on the image versus a reflectance, how much of light emits back from the object when it is being illuminated. There are slight differences in X-ray and medical images. You don't in X-ray light is not reflected. Light passes through an object and you form an object on a sensors placed behind the X-ray and hence reflectance will be replaced by transmittivity. So some bones won't allow light X-rays to pass through, whereas soft tissue will allow the light to pass through.

And hence you can see where are the regions blocked by X-ray and those then can be inverted to see the bones in a bright color. So we finish so far with session two of week one.

Want the rest of Advanced Medical Image Processing in Clinical Applications?

Get the full course syllabus and fees.

The full course is paid (₹40,000). We'll send the syllabus and call you back within one working day.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Full course · paid

Unlock the full Advanced Medical Image Processing in Clinical Applications course

The remaining 42 lessons, mentor-reviewed projects and a certificate come with the paid course (₹40,000). Leave your details: we'll send the syllabus and fees and call you back.

Get the syllabus and fees

A counsellor calls you back within one working day.

By submitting you agree to our Terms and Privacy Policy.

Grow by Skill-Lync · free lesson

Start watching

Free, no payment. Leave your details once and every free lesson on Grow plays straight away. A Skill-Lync counsellor may call you about the full Advanced Medical Image Processing in Clinical Applications course.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.