XnView Software

First of all: Great that such a feature is now in XnView 1.9 alpha and morover: It works quite good for my test files. BTW: Message comes from thread http://newsgroup.xnview.com/viewtopic.php?p=36720#36720

As you asked for suggestions, here are some:

• Please let me choose which attributes to display, just like for the browser, so I have access to what information is important for my task (eg EXIF)
• rotated pics (eg 3°) are recognized as very different (eg 75)
• drop shadows - with elsewise same content etc. - provoke huge difference (in my example 14 points!)

Some time ago I tested several image similarity programs. Here is what I jotted down and what mit serve as inspiration. From ImageDupeless 1.6.3 http://www.imagedupeless.com/en/

• comparision of 2 files with big pictures (=see the details) and thumbails of their direct neighbours (very helpful for picture series).
• provides hint, what pic the program thinks is of better quality (so which one to keep). Works usually quite good as it is "intelligent", so for binary identical files the shorter filename is worse, the newer file is worse (as the older one has usually the original date), for different files the smaller ones but depending on compression ratio,...

Inspirations from d'peg! 6.17 http://www.gotdupes.com/

• Highlight what is different between two similar images (size, color depth, compression ratio,...).
• offer (semi-)automated tasks like move, replace, rename,... based on results
• let me choose the comaprision methods used (histogram, SVMs,...) and their sensibilities

For further inspirations, you may have a look at the (not as good as the above ones but still competing) programs Retriever Image Viewer, Thumbnails & Database 2.3 http://www.djuga.net/retriever.html and PhotoSort v1.10 http://www.asmdev.net/products/photosort/.

/Georg

XnView 1.90

GeorgD wrote:As you asked for suggestions, here are some:

• Please let me choose which attributes to display, just like for the browser, so I have access to what information is important for my task (eg EXIF)
• rotated pics (eg 3°) are recognized as very different (eg 75)
• drop shadows - with elsewise same content etc. - provoke huge difference (in my example 14 points!)

Other suggestions:
- better handle images with variations in Luminosity/Contrast...
- find images that have small parts with high similarity (instead of using full image comparison only) => find cropped versions, copied/pasted elements, composited images, etc...

All this requires longer discussion & implementation, though -> Postponed.