Higher Resolution images with low pixel data

Hi,

I am looking for a higher resolution image of a detector data. The detector is a high energy detector having 44 x 44 pixels and it is not physically possible to increase the number of pixels. By saying 'higher resolution', I do not intend to mean higher optical resolution, but what I essentially need is to convert this low pixel data into a higher pixel data. You can see a checkerboard pattern in the below image (Left), which in effect preserves the details at the region defined by the pixel intact -- which is good technically, but doesn't give a nice outlook to the image.

I did try using imresize, these options seem to smudge the data over a wider region and hence, distort the image. Also, There seems to be the presence of 'Halo' or glows around the object. What I need is a technique which preserves the localization of data to the regions, yet produce a higher dimentional image.

Would be of great help if someone shares a solution to this issue.

Regards

Murali

Answer 1

[Walter Roberson]

No.

• replication of pixels does not introduce new data, but it does not look good
• interpolation does not introduce new data, but it smears what is there, in a way that can sometimes be misleading
• In some cases, when you have a model of the underlying process, you might be able to fit the model to the detected locations and then use the fitted model to reconstruct a nice looking image. Whether localization is preserved depends upon the model and how easy it is to separate the regions. For example it is notoriously difficult to get good fittings for overlapped gaussian curves that have phases that have to be fit.

[Murali]

Hi Walter

Thank you for your answer.

The underlying process follows poisson statistics.

Do you have any material/literature to go ahead with what you suggested?

Thanks

[Walter Roberson]

At the moment, it looks plausible to me that you have isolated points that are too far away to interact, and that at each location, you have a radially symmetric process. If that is true, then you can do region finding to find the blobs, regional max to find the centers, calculate distance of the pixels relative to the centers, and fit the poisson statistics based on radius. Though likely it would be better to also fit the exact position of the center.