There is one very suitable sentence which can often be seen at the beginning of the scientific paper (and, I admit, at the beginning of the “Image restoration” Wiki page) as an answer to the question from the title. This sentence briefly explains that: “Image restoration is the operation of taking a corrupt/noisy image and estimating the clean, original image” [1]. For all of you who are not reading scientific papers on this topic, let’s decompose these sentence.
What is a corrupted/noisy image?
Noise is the word most often used to describe the unpleasant sound, which is not wanted by the listener, like the sound of the hungry can scratching on the fridge door. In the case of the image, noise is an unwanted corruption of the image – variation of brightness or colour information in the images which we do not want to see, usually caused by the equipment used for taking the image. The second important type of corruption is a blur. Imagine that you are a bird lover why spotted a beautiful, rare bird on the branch and at the same moment when you are ready to take a picture, bird take off and instead of the image of this beautiful bird you have some smeared structure which, in other people eyes, can be domestic cat jumping from the tree. For this kind of images, we say that they are blurred.
What is a clean image?
This question seems easy, but do not be deceived. Maybe you will say that clean image is one which is more pleasing to the observer. This is half true. The other and the most important half is that the clean image, in the case of image restoration task, needs to represent a realistic data (ground truth). If we have as an outcome more pleasing image, but not necessary “true” one, we are talking about the image enhancement. You can imagine that in some applications, like in medical imaging, we need to be particularly careful not to lose any information from the clean image.
What is estimation?
Estimation is the process of finding an approximation - value that is usable for some purpose even if input data (image in our case) may be incomplete, uncertain, or unstable. We are using the best available information and some previous knowledge about the image (structure, statistic and so on) and equipment.
So, to conclude, here we have an inverse process: we are starting from the corrupted image, trying to reverse corruption to get a clean image which is as close as possible to the true one. At the end, maybe, our resulting image will not be perfect, but we can convince our friends that we saw Forest Owlet (one of the rarest bird in the world) and not neighbor’s cat.
 |
| Clear and blurred image of the owl |
This comment has been removed by the author.
ReplyDeletewell explained
ReplyDelete