On Blurriness

My claim is that when you see X sharper i.e., when you see a sharper image of X, you see more of X (in it).

In the literature (for example, Tsomko, Kim & Izquierdo 2009), blurriness is seen as a property of pictures, resulting from a variety of causes: lack of focus, movement of objects, movement on the part of the subject, distance, ambiental obscurity, etc. There are two ways of identifying blurriness in images: a semantic and a syntactic one. From a semantic perspective, and regardless of the cause, blurriness/sharpness can be understood in terms of the quantity of information contained in the pictures, i.e., how much about the picture’s subject you can obtain from the picture.

The basic idea behind the semantic approach is that there are properties of different granularity. For example, if you see someone from far away, you may be able to see that she has eyes but not what color her eyes are. If you directly see a racing car on a speedway, no matter how up close, you might not be able to see its number or, say, whether the driver was smiling as he passed along or not. In other words, there is stuff that you cannot see when an image is blurry. Thus, when your image off something X is blurry, this entails that there is some stuff of X that you would otherwise be able to see if your image of it was sharper.

From this perspective, having a sharp image of X is not seeing X in a different way that having a blurry image of it, but seeing facts about X (i.e. X seeing X as having certain determinate properties) that X objectively has that you would miss when your view of X is blurrier. This precludes the possibility that a blurry and a sharp image contain the same information (i.e. that they both represent the same object as having the same properties). However, it does not preclude the possibility that a blurry and a sharp image represent their objects as having the same property. To better understand how this is possible, it is best to take a closer look at the second, syntactic way of understanding blurriness.

From a syntactic perspective (useful when you cannot compare the picture with what it represents), the most usual technique for identifying blurriness is by the absence of sharp edges (Marziliano et al). Thus, an image is blurrier than another if it contains less ‘spiky’ differences, say, if the average difference between adjacent regions in the picture is low (well, actually there is a whole interesting debate in AI as to how to formally define edges and sharp edges in particular. See, for example, Elena Tsomko,  & Hyoung Joong Kim 2008 or Tsomko, Kim & Izquierdo 2009).

Now we can see how a blurry and a sharp image can represent their respective objects as having the same properties (and thus, be visually indistinguishable), i.e., if there are two objects A and B which differ only on the softness of their edges (say A has sharper edges than B), then there must be some level of blurriness at which looking at A must look just like looking at B.

It is very important to notice that the semantic conception is the proper understanding of blurriness and that the syntactic conception was developed as a defeasible mechanism to detect proper blurriness.

The problem with the Gabbor patch is that it is not a picture of anything, as such, it is odd to say that it is a blurry image. I mean, if it is a blurry image, what is it a blurry image of? Instead, the proper thing to say would be to say that it is an object with soft edges. In other words, even if it seems to satisfy the syntactic notion of blurriness (it has no sharp edges), it does not satisfy the semantic (and more substantial) one.

Originally published at https://www.philpercs.com/2019/08/what-is-blurrinesssharpness.html  on 08/22/2019 11:59:14 AM