# point of view of each pixel with regard to the whole image

I am looking for a technique or an algorithm that outputs a unique number/vector for each pixel of the image. I want to give the pixel a perspective so the neighboring pixels have more visibility and the farther away pixel have less visibility. Imagine if someone walks on the pixels, he sees different patterns, and I want to get such pattern for each pixel. I tried to think of the 2D image as a 3D image stretched on a sphere, but then I would miss the neighboring pixels, and I also thought of planar convolution, but I had no luck, or very intuitive method. I hope that I could explain what I am looking for. Please ask your questions so I could make it more clear.

• Please clarify what your requirements are. Assigning each pixel the value $(x,y)$, where $(x,y)$ are its coordinates, gives a unique vector to each pixel, but I doubt you'll be satisfied with that, which makes me suspect there are some requirements missing. I don't know what it means to give a pixel a perspective, or what it means for a pixel to have more visibility, or what it means to miss neighboring pixels. Can you formulate your requirements precisely? Perhaps it might also help to tell us your motivation or the context surrounding this question.
– D.W.
Feb 12 at 20:00
• Thank you for the comment. I try to explain it in a better way: Final goal: profile each pixel of an image, based on its neighbors. the close the neighbors are, the more important they are in representing the pixel. We could use (i,j), but I want to use the pattern of image itself, and not the 2D coordinates. I hope that it clears things up in abstract level. I will try to explain my initial question in the next comment. Feb 14 at 10:57
• Perspective of a pixel: a pixel has a 2D location in an image (MxN). If we consider pixel in (0,0), then if we give the pixel a height, e.g. a person hypothetically stands on the pixel itself, he will see the pixel that has more distance from it smaller, and has wider field of view, and he sees the neighboring pixels larger. I want to use this concept as an intuitional way to profile the pixel itself. Each pixels has its own view of the image, and it is unique in its own way, but the thing is the neighboring pixels have similar views, and I want to use this. Is there a better way for that? Feb 14 at 10:59
• Sorry, I still don't understand what you want the output of the algorithm to be. I don't know what it means to "profile a pixel" - what should an algorithm output? You say a "better way for that" but I don't know what the "that" is. Can you describe what you want the algorithm to output precisely?
– D.W.
Feb 14 at 19:04