Hello and welcome back. It's hard to believe we are already done with one third of the course. One of your classmates suggested in the forum that I welcome you once, or at least once, I guess, in Greek So, [FOREIGN]. This is going to be a fun week. Image and video enhancement is a fun topic. One of the challenging parts of the course is over and you might say that it will be smooth sailing moving forward. While some topics will be more mathematical than others. Most of the topics that we'll follow are application driven, and probably not as demanding as the material of weeks two and three. Enhancement is a fun topic because we will lend certain tools that we can utilize right away, and which will allow us to play, in some sense, with the images and videos. On top of that, we will be the judges of how good and pleasing their enhanced image is. So, based on the tools we learn this week, we can carry out and try your favorite enhancement technique on an image. And if you're not happy with the result, you can change your field and parameters, which control the performance of the technique. Or just try another competing enhancement technique altogether until you are satisfied with the results. It's a fun topic because it's very intuitive, and rather straight forward to understand, or at least the better part of it. Enhancement is certainly one of those processes that it is used everywhere at all times. You don't like how an image looks, well, let's enhance it. Actually, all image processing software packages include a good number of enhancement modules in their repertoire. So, during this week we'll introduce the topics of enhancement, and distinguish it from the topic of restoration. Which you will cover during weeks six and seven. We will then first cover, point wise information of enhancement of an image. Either based on a general objective. Such as stretching the contours of an image. Or by analyzing the image and, more specifically, its histogram. Imposing an objective on the histogram, such as equalizing it. And then deriving the pointwise transformations of the intensity values. After that we'll cover transformations of the intensity value of the image based on the properties of the neighborhood of a pixel, performed both by an LSI system that we discussed in weeks two and three, but also by a nonlinear filter. We will then cover techniques which allow the image to look sharper. A specific a technique for reducing the dynamic range, while at the same time, increasing the local contrast of an image through homomorphic filtering. The notion of adding color to a black and white image, and finally some simple notions of enhancing a video sequence. So, let us start this exciting week with an introductory segment where the distinction between image enhancement and recovery is made. We will cover this week image enhancement while in the following weeks, we'll be covering image recovery. It is therefore worthwhile drawing the distinction between these two related topics. With enhancement the objective is to manipulate or transform an image so that its usefulness or visual appearance is increased. No explicit determination is made of how this increase in usefulness or visual appearance will be measured. With restoration, or recovery, we are modeling the degradation the image is subjected to. And our objective is to remove this degradation, and therefore recover some of the information that was lost, based on an optimality criterion. So the two important steps that are involved, the modeling, and the determination of an optimality criterion based on which this degradation is removed. Let us now look at some examples. The first one is the modification of the intensity values of an image so that the contrast is increased. Since no modeling of the degradation is involved, at least explicitly, and no optimality criterion is determined based on which the contrast will be increased. This is definitely an enhancement approach. The second example is deconvolution. In this problem, the degradation is modeled as the convolution of the original image with the impulse response of the degradation system. So in other words, the degradation system is linear and spatially invariant as we've learned earlier. And the objective is to undo this effect of the degradation system or deconform the image based on an optimal criterion. So this is therefore certainly a restoration problem. The third example is inpainting. As you'll see the degradation in this case removes the intensity values of the pixels in the part of the image, and the objective is to estimate these missing values. So, if this estimation is done without determining an optimality criterion, just doing an operation that simply makes sense, then we have an enhancement problem. If on the other hand the impending procedure depends on an optimality criterion then we have a recovery problem. So this could be either enhancement or recovery. In this we'll present talk about inpainting as a recovery problem. And the final example is the smoothing or removal of noise in an image. This problem also can be either an enhancement or a restoration problem. Depending on whether the noise characteristics are modeled, and also whether a specific optimality criterion is defined. We'll cover actually noise smoothing in this class both as an enhancement and as a restoration problem. It should be noted here that this distinction between enhancement and restoration is here to guide our presentation of the various approaches. And also to guide our thought process when we're trying to solve an image processing problem. It's not here to somehow erect solid walls between these two topics and somehow penalize somebody if an enhancement technique is called a restoration technique or the other way around. Here is a summary of some of the main characteristics of image enhancement algorithms. There are no theoretical guidelines in approaching an enhancement problem. Each enhancement problem is different and the criteria for enhancement are typically subjective. We want to increase the visual quality of an image, for example, but each viewer typically has a different quality norm in their mind or in their eyes. Each of us most probably will have a different preference when we are presented with processed versions of an image and we are asked to find which one has the best quality. This criteria might even be too complex to convert them to useful objective criteria. Because of that, the enhancement algorithms are qualitative and ad hoc. They kind of make sense and in most cases they work well, but they're also application-dependent. A different transformation of intensities might be performed if a user is to evaluate the image, or if this image is input to an object detection or an object tracking or a classification system. Finally, coupled with the previous comments, the evaluation of the effectiveness of the enhancement algorithm should also be application dependent. Again, if the enhanced image is intended for a human viewer or for a classifier, the criteria for evaluating its effectiveness should differ. So enhancement techniques in general are easy to understand, and they've been very useful and very widely used. And the large number of them are actually implemented in software packages. So we will present the number of techniques next and show examples where, after enhancing, the image is more pleasing to look at. Or it provides additional information to a classifier, or an object recognition of object tracking system. We'll also see examples where features or objects in an image that were not visible before enhancement are now visible after enhancement. So let us now look at some of these enhancement approaches. The image enhancement topics we'll be covering in this class are Point-wise Intensity Transformations that are defined ahead of time to suit our purposes towards enhancing an image. Point-wise Intensity Transformations that result from objectives we impose on the histogram of the image. Spatial filtering, both linear and spatial invariant and non-linear. We'll see examples for filtering for noise smoothing, sharpening, homomorphic filtering, applications of pseudo-coloring, and some examples on video enhancement. Let us proceed now with the first topic of the list.
