Sabtu, 23 November 2013
These days I find my brain battling with the differentiation of which brain hemisphere to admire, upon glancing at a picture in a media stream. Is credit to be attributed to someone creating a beautiful, artistic picture that manifests an underlying creative human mind, or is credit due for a more rational mind capturing the ultimate marvels of nature[1]. The sort of impeccable beauty, frozen in time by a camera, and post-processed in a streaming digital glory of ones and zeros. Ethereal captures like these:
Mending confusion over which is which, that is if the photo in question is an artwork or real, would require properly cited sources, which is often rare to come by in aggregated social media streams. Frankly it may even be much to ask for, as people post for different reasons, and follow different ideologies. Yet it stands to reason that many people do not wish to miss mind-bending new impressions of their natural world. Few among many would actually like being inadvertently misled, when a picture later turns out to be a work of fiction. In the worst case, it may even backfire on the artists, whose work was reproduced uncited and with different intent. We are curious creatures in nature, and care about getting to know the limits of our natural world. Who could blame us, with a marvelous world as ours. The APEX rules, short for Author, Pristine, Enhanced, X-reference or crossreference are aiming to quickly differentiate a picture when its origin is cast in doubt.
The APEX Test
As it turns out, a few basic steps requiring only seconds of a user's-attention, are sufficient to roughly qualify if a photo is a capture or artwork:
Embed / Share
I posted these guidelines as a gist to github. To embed it simply copy and paste the following HTML into your post, or article.
An incredible example
As on of the best comments among hundreds, +Andrel Rivers points out that:
A credible example
Here is a interesting photo, shared by Armin Ronacher, that would checks out as follows:
Possible Future as an Extension
It would be interesting to develop the idea of image-credibility-checks into an fully automated script, along with a user rating database. The look would be similar to:
Hoovering for a second over non-credible images would highlight all like-rated on a given page, likewise, hoovering over all credible images would highlight all those classified as credible, such as with a rating of at least 4 stars out of 5.
Scientific Background: Why APEX seemingly works
When doing a binary test, such as in this case classifying between credible/real and non credible/artistic several significant measures have to be regarded. Sensitivity or true positive rate, which is the number of verified positives divided by the of verified positives and missed positives and specificity, which is the same as before in reverse, are paramount to make sense of a test's performance.
For now APEX has only been tested roughly and subjectively on a small scale. As such the ability to report concrete performance measures is futile at the moment.
The effects that make the four APEX rules apparently sufficient are likely complex, but at least two effects may be attributable:
User Questions
Footnotes:
[1] The left / rightside hemispheric brain model is likely obsolete. Music, creativity and the likes are likely global hemispheric phenomena.
 |
| Austrian mountain dipped in a sea of fog, by +Stefan Brenner |
Mending confusion over which is which, that is if the photo in question is an artwork or real, would require properly cited sources, which is often rare to come by in aggregated social media streams. Frankly it may even be much to ask for, as people post for different reasons, and follow different ideologies. Yet it stands to reason that many people do not wish to miss mind-bending new impressions of their natural world. Few among many would actually like being inadvertently misled, when a picture later turns out to be a work of fiction. In the worst case, it may even backfire on the artists, whose work was reproduced uncited and with different intent. We are curious creatures in nature, and care about getting to know the limits of our natural world. Who could blame us, with a marvelous world as ours. The APEX rules, short for Author, Pristine, Enhanced, X-reference or crossreference are aiming to quickly differentiate a picture when its origin is cast in doubt.
The APEX Test
As it turns out, a few basic steps requiring only seconds of a user's-attention, are sufficient to roughly qualify if a photo is a capture or artwork:
 |
| Source: wikimedia, License: CC-BY-SA 3, Author: Larry Yuma |
- Author: Check the source by looking at the stream of the original poster (OP), conveying intent.
- Pristine: Does the image have at least several mexapixels or is it greatly reduced? Does the file have metadata such as Exif data?
Note: A normal degree of data erosion is normal. After all, reposting and retweeting is a bit like a message in a bottle. - Enhanced: Is it sensible what you are seeing, and at in some physical agreement with your common sense or gut feeling. Simply put, does your first impression make sense to you?
- X-Reference: Start an image cross-reference search. TinEye / Facebook offers a great engine to do just that. Here is an exemplary search ( picture source). Be patient, though as Tineye are in the process of building their database.
Embed / Share
I posted these guidelines as a gist to github. To embed it simply copy and paste the following HTML into your post, or article.
<script src="https://gist.github.com/lsauer/7614189.js"></script>
An incredible example
As on of the best comments among hundreds, +Andrel Rivers points out that:
The clouds are layered. At the top, u notice the smoke going into the cloud. Then u see lightning in two different colors/shades. Along with the cloud being wrapped in lightning which is scientifically and physiologically impossible in nature. I was a computer graphic designer and taught photo shop for a few years. Along with living in Volcano Hawaii/ Big Island. The only natural element in this photo is the mountains in the back drop. In fact, that entire right portion of this image is photo shopped. U can also see the outlines if you know what to look for.
So like I and I said, photo shopped, but nice image regardless. Super imposition is all it is.
A credible example
Here is a interesting photo, shared by Armin Ronacher, that would checks out as follows:
- Author: At a quick glance the stream would appear overall credible, with an informative intent✓
- Pristine: Full sized image and Exif information✓. No disturbance or smudging are in the image background noise pattern.✓
- Enhanced: The image appears unaltered, not smudged, and shows expected camera limitations such as a limited dynamic range, and color fringing. The setting is mundane. ✓
- X-reference: Empty, Not required if you know the depicted context from personal experience
Possible Future as an Extension
It would be interesting to develop the idea of image-credibility-checks into an fully automated script, along with a user rating database. The look would be similar to:
 |
| A picture example overlaying a credibility-rating view via a browser extension |
Hoovering for a second over non-credible images would highlight all like-rated on a given page, likewise, hoovering over all credible images would highlight all those classified as credible, such as with a rating of at least 4 stars out of 5.
Scientific Background: Why APEX seemingly works
When doing a binary test, such as in this case classifying between credible/real and non credible/artistic several significant measures have to be regarded. Sensitivity or true positive rate, which is the number of verified positives divided by the of verified positives and missed positives and specificity, which is the same as before in reverse, are paramount to make sense of a test's performance.
For now APEX has only been tested roughly and subjectively on a small scale. As such the ability to report concrete performance measures is futile at the moment.
The effects that make the four APEX rules apparently sufficient are likely complex, but at least two effects may be attributable:
- A dilution effect of few professional picture-fakes by high numbers of low-resolution picture-fakes. The reasoning may be found in the technical advancements: Whilst it still takes skill and experience to produce high resolution fakes, is has become very easy to create low-resolution fakes, owing to advanced algorithms for retouching, distorting and relighting images. In many cases the results will look good at a fraction of the original pixel count. As our world is fractal in nature downsampling of a picture likely leads to new quality of perception. What would have been formerly perceived as a retouching blemish or smudge in a high resolution image, may then be qualified by our visual perception system as a feature.
- The second significant effect occurs in the APEX rule, Author, upon looking for user intent of the original poster (OP), such as posting pictures indiscriminately or already accruing content of either a real or artistic variety. In the non-discriminate case, essentially a pre-qualification step is undertaken by a human classifier rather than an computer algorithm. As such in the worst case one classification step occurs by APEX, at best two classifiers are combined. One classification step through APEX and the second step by recognizing the original intent of the poster.
User Questions
- Do you see a need for an extension?
- Do the APEX rules sufficiently work for you?
- Do you have caveats?
Footnotes:
[1] The left / rightside hemispheric brain model is likely obsolete. Music, creativity and the likes are likely global hemispheric phenomena.