Image Find (iFind) Help
 
iFind is a free Win32 command-line tool to find images. iFind can be used in two ways:
- to find an given image in a set of images
- to find similar image groups in a set of images.
Similarity Coefficient
iFind relies on the correlation coefficient between to images to figure our their similarity. The similarity coefficient takes values from 0.0 to 1.0, where 0.0 means no similarity at all, and 1.0 means the images are the same. The default similarity coefficient used to compare images is 0.60, which should be ok for most cases. The similarity coefficient can be changes using the -s option.
Finding an Image
iFind is useful for example if you scan a printed photo and want to find which original image from your photo collection was used to print the photo. The image to look for can be resized, have noise, have color transformations, etc.
To compare a given image a.jpg with another image b.gif use:
<![CDATA[
ifind -i a.jpg -d b.gif
]]>
To compare a given image a.jpg with all images in folder c:\my images\ and in all its sub-folders, with a similarity coefficient of 0.5, use:
<![CDATA[
ifind -i a.jpg -d "c:\my images\" -s 0.5
]]>
The found images, if any, who pass the 0.5 similarity boundary will be printed in the screen (in the standard error), along with their similarity.
In the case you are interested only on the first two matches, use:
<![CDATA[
ifind -i a.jpg -d "c:\my images\" -s 0.5 -c 2
]]>
A fast, but less accurate search is possible if you append the -f option. By default, iFind reports a match only if one color of the images matches. To force a match all colors append -a:
<![CDATA[
ifind -i a.jpg -d "c:\my images\" -s 0.5 -c 2 -f -a
]]>
If you specify the -h option and some images are found, then a HTML report is generated and open in the system browser:
<![CDATA[
ifind -i a.jpg -d "c:\my images\" -s 0.5 -h test
]]>
The HTML report file is named test-ifind.html.
Finding Similar Image Groups
iFind can find similar image groups in a given set of images using their inter correlation. The grouping is based on correlation and this means the groups will contain the same of very similar images. It does not warranty that the images of a given group belong to the same logical set and vice-versa. Only the repeated (modified) images can be found.
This kind of correlation based search, while possible, requires around N * (N - 1) / 2 comparisons between N images, so it has a O(N2) complexity. Grouping done by ifind tool is thus not feasible if you apply it over a large number of files!
To find possible similar image groups, if any, with a given similarity of 0.3 in the folder c:\ my images and all its sub-folders use:
<![CDATA[
ifind -g "c:\my images\" -s 0.3 -f
]]>
The options -s, -f and -a. -h are valid with -g and have the same meaning as with the single image search. The other options are ignored. Usually using -f is a good idea to speed up the comparison.
Usually with grouping you should be a bit more tolerant on the similarity coeficient -s value (use smaller values) than with the single image search.
The files to compare are first scanned and loaded in memory, and then they are compared. The number of comparisons needed is printed. The found groups if any are printed in the end along with the files of each group.
To generate a HTML report use the -h option:
<![CDATA[
ifind -g "c:\my images\" -s 0.3 -f -h testreport
]]>
The HTML report file is named testreport-ifind.html. The file links directly the images and may take some time to show in the system browser.
|
