|
|
Post by FWDekker on Aug 27, 2014 8:00:42 GMT
Before I suggest an idea that may already have been implemented, I'd like to ask what the current state of the image comparison is. When an image is (attempted to be) posted, how is it exactly compared, hashed and saved?
|
|
|
|
Post by GandalftheGray on Aug 27, 2014 11:19:13 GMT
Before I suggest an idea that may already have been implemented, I'd like to ask what the current state of the image comparison is. When an image is (attempted to be) posted, how is it exactly compared, hashed and saved? An algorythm compares MD5 values. It doesnt always work tho. |
|
|
|
Post by FWDekker on Aug 28, 2014 8:42:39 GMT
Before I suggest an idea that may already have been implemented, I'd like to ask what the current state of the image comparison is. When an image is (attempted to be) posted, how is it exactly compared, hashed and saved? An algorythm compares MD5 values. It doesnt always work tho. If it is simply the comparing the MD5 values of the images, there will only be a hit if the images are exactly the same (with the exception of collision). Changing one pixel would completely change the hash value. Wouldn't it be a better idea to implement keypoint extraction so that the features of the images are saved, so that later there will also be a match if the images are merely similar? |
|
JRX
New Member
I really like Potatoes
Posts: 23 
|
Post by JRX on Aug 29, 2014 22:32:55 GMT
That would finally make the "Repost Search" work properly, but i guess it's harder than it looks
|
|
|
|
Post by emphram on Oct 17, 2014 21:58:27 GMT
There are several complications in regards to this. I could imagine an algorithm that averages out an image to a certain resolution and then compares it with another image also averaged out to that resolution and flags any two images within a certain range of similiarity, but then this would have the problem of flagging memes with different text and would only become slower and slower and more resource intensive as content accumulates, even if a queue system were to be implemented, and a user had to wait x hours before his or her post were published (and the queue would only get longer and longer). No easy way and efficent way to implement this.
|
|
