StackOverflow2013

Note that there are some explanatory texts on larger screens.

plurals

PO
text
Body
copied!<p>The simplest thing to try is to compare the <a href="http://en.wikipedia.org/wiki/Variance" rel="nofollow noreferrer">variances</a> (standard deviations of the image) -- since the image content is almost the same in all frames, the sharpest image will probably have the highest variance. Variance of the inter-frame prediction residual (differences between adjacent pixels) may also work -- look around for <a href="http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=4556912" rel="nofollow noreferrer">this paper</a>.</p> <p>Here are two images. An original is on the left. I simulated motion blur by Gaussian blur in the horizontal direction only. The blurred image is on the right.</p> <p><img src="https://i.stack.imgur.com/vcTiL.png" alt="original"> <img src="https://i.stack.imgur.com/HViLb.png" alt="blurred"></p> <p>Standard deviations (using <a href="http://rsbweb.nih.gov/ij/" rel="nofollow noreferrer">imagej</a>) : 42.5 and 41.1</p> <p>If that doesn't work, then what you're after is called <strong>motion blur estimation</strong> (or just blur estimation in general). It's a pretty popular image quality assessment problem in academic circles, so there's definitely lots of material to help you. </p> <p>Paul R got you started on the right track -- look at the edges in the image. </p> <p>It's not necessarily the number of the edges that matters here -- it's their <strong>width</strong>. Blurry images will have wider edges (more distance between the rising and falling side). Read <a href="http://stefan.winklerbros.net/Publications/icip2002.pdf" rel="nofollow noreferrer">this paper</a> -- it's very popular, and quite simple to implement. Note that that approach only measures blur horizontally and vertically, which wouldn't work in your case. However, it isn't hard to modify it taking <em>edge direction</em> into account, as some later papers have done.</p> <p>There are more ways to deal with this problem, but the ones I've mentioned are the simpler ones that I know.</p>

Querying!

Guidance

An individual column

Larger individual text columns get their own page to allow for proper reading.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload