I recently cleared all of my images off of my phone, and ended up with hundreds of duplicate images on my hard drive (due to the fact that I had downloaded/synced most of these images in months past). I needed a simple and straightforward duplicate image finder, and found one in a freeware called ‘Similar Image Finder’ .
‘Similar Image Finder’ is very easy to use, and revolves around a single number: a metric called ‘minimum similarity’ which is roughly the percentage of ‘exactness’ that you wish to use. This number can range between 100% (for an exact unresized, unaltered duplicate) and will let you can experiment with values in the middle to identify images that ‘look like’ each other. However, in our test of this software the ‘minimum similarity’ number loses its usefulness pretty quickly at just under 90%, where the two paired images might as well be paired at random.
Sorting results: you can click a column header to sort by it, but somehow cannot add ‘size of file’ as a column. This means that while you can sort by the ‘similarity’ score, you cannot sort by file size, which is what most people who want to remove hard drive hogging duplicate images would be interested in doing. A rather strange omission.
But the program will let you determine which is the original and which is duplicate based on your choice of: file size, creation date, or pixel size.
You can shift-select bunches of files to perform operations on them,
The verdict:
If you are looking for an image deduplication program that is easy to use(and free), then you will like this program. There are other image de-duplication freeware that have more options (esp. options for filtering of results), and might offer better results, but they are also more involved and have a bigger learning curve.
Users who are looking to get rid of exact duplicate files can set the figure at 100% and be done with it, while others who want a more sophisticated de-duplication scan involving altered or visually similar images can place less exacting criteria and experiment with the results.
Using a single quality criteria percentage is, in theory, a very useful and user friendly concept. However, in my experience, the ‘similarity number’ becomes meaningless (and rather useless) just under 90%, where the two paired up images can be very different and still be reported as potential duplicates (see the example in the screenshot below, which came in at 87% similarity). The algorithm seems particularly vulnerable to flag blank (mostly white or mostly black) images as duplicates for all sorts of images erroneously.
Wish list: this program can be easily improved. If the algorithm would be careful around mostly blank images, the results can be much better. Also, if other criteria could be added as columns to filter by, such as size or others. The option to CTRL-selecting individual items would also be a great help to users (as opposed to the current shift-selecting only).
Go here to download Similar Image Finder (Windows).
