Image Investigator
Description
1. Overview
Our new developed program is a powerful image investigator. Our Image Investigator can scan large image archives and retriev or identify similar images, which may differ in size and format. This function can be used to scan image archives for duplicates or for images. by similarity. Images can also be located by using a simple drawing or color pattern.
Our Image Investigator is a useful tool for image agencies, news paper companies etc. These companies rely on text labels or persons with a good image memory, which is unreliable and images may not be identified. Our program can help to find images to identify copyright infringements, modifications or other misuse. It is also very useful for professional webmaster, photographers, designers or other person, who deal with a large number of images. Our program does fast multiresolution image queries or image retrievals from large databases.
2. Introduction
Of course, images to be scanned have to be inspected by our program first. This can be a long lasting procedure when many images are involved. To make this less dramatic, we have implemented many, here un-described, tricks. One of our trick is to gather what we call finger prints of the images. These are used in the actual compare, which allows to have also a million images in the archive. To make the job even faster, finger prints can be saved to hard disk and therefor re-used at a later time. We just have to check for changes and the search can start. The number of Archives is limited in the demo version.
The actual compare we do accelerate by several other tricks, like checking for an adjustable aspect ratio tolerance. This avoids compares of portrait images with those in landscape format. Further on many performance controlling settings are available. Especially, the sample size or finger print size is of large impact on performance, the number of images the program can handle and the memory consumption. The control display, which show images being processed slows the compare by a factor 3.
In case, many images are like just gray and therefor very similar, a match limit per image can avoid negative effects.
3. Performance
The program has been tested with many thousand images. It is very fast, but no absolute statement about the speed can be made. The performance depends not only on the computer being used, but also on the complexity and size of the images. Comments to the settings can be found in Forms and Menus. For the computer hardware, the amount of installed memory is of great impact. If a lot of memory is being used by the program, the operating system will start to swap parts of the memory to the slower hard drive. This can degrade performance significantly. We try to avoid this, by using our own intelligent disk mode, if we cannot reserve enough RAM (memory) for our finger prints. This is a little slower, but acceptable, if not enough memory is available.
Some figures for an archive of 42000 images (4.22GB), a finger print size of 48, a threshold of 75% similarity and using 2.6 GHz Pentium 4 PC.
- Inspecting images and creating finger prints: 18 minutes
- Start and reading Disk-Cache: 4 seconds. (unmodified archive)
- Start and reading Disk-Cache: 12 seconds. (Archive contains 80 new images; incl. saving cache to disk)
- Compare of 30 unbuffered images against the archive and active program: 5 seconds. (Folder dragged to on active program)
- Compare of 30 unbuffered images against the archive and in-active program: 7 seconds. (Folder dragged to on the program icon)
- Scanning Archive for duplicates 7238 seconds with 438 matches. (every match costs additional time; Disk Cache was read; scanning 42000 images for duplicates are 882000000 tests, this is approx, 122000 Tests per second.
3. Limitations
This program compares images, not image content! It doesn't recognize persons in different images as that is content.
The total number of images, our program can handle, is limited by a 6 Giga Byte barrier. Almost 1 Million per dataset can be handled for a sample size of 48 and approximately 3.7 Million for a sample size of 24.
Our Image Investigator supports JPEG, JPG and BMP only at first. TIFF, GIF, PNG, CUT, DCM, DNG, FAX, FITS, JNG, MAT, JP2, JPC, MPC, MVG, PALM, PAM, PCDS, PCL, PCX, PDB, PGM, PNM, PPM, PSD, PTIF, PWP, RGB, RLA, RLE, SCT, SFW, SGI, PWP, SUN, TGA, TIM, UYVY, VICAR, VIFF, WBMP, WPG, XCF, XWD, XCBCR, XCBCRA, YUV and EPS are turned off by default.
Gray landscape pictures are not well suited for our program.
Forms and Menus
More Forms in A-Z and Menus when right mouse button is clicked.
Results
If images match the defined criteria, they will be shown in the list (above). The List gives information which files are similar in %. If a match is found by mirrowing one image a "M" next to % will indicate this. In addition mini images of the two images involved and a difference images (XOR) are displayed. The Investigator Dialog can be opened per right mouse button or double click to get a large view. There are also other functions as Open Folder, Sort Results, Help and Menus to add Results to the Ignore List or delete them form the results.
Basics Settings
This form contains settings, which can be modified during a scan. Changes effect only future compares.
Closest matches are rough estimates only and might appear randomly selected when values are very similar. NOTE: the background will turn yellow.
Advanced Settings
Changes on Sample Size will cause all images to be read again to generate new finger prints.
Show Frequency Distribution has no effect on a running search. Frequency Distribution
"Don't verify folder automatically" turns of automatic verifification of folders, as reading folders with many files can take a long time. Look into Maintenance to initiated verification manually.
Save "Cache Buffer" saves generated finger prints to the hard drive and allows to for fast startup times. This number of Archive is limited in the Demo Version.
Log
If stated in the scan result message, more information about a scan will be available in this form.
Archive Maintenance
Dialog to manage the different data sets.
An archive will be added, when its data incl. finger prints have been saved to disk. Every archive will get a name automatically, which can be re-defined by the user. This form can also be used to select archives, if checked in Advanced Settings. It will be shown then when Archive is clicked.
"Verify" forces a cache verify before the next search. This is to have the cache up-to-date. Normally the program checkes for folder changes, but depending on the underlying file system, changes might be recognized only when a data set is opened.
Frequency Distribution
Frequency Distribution gives you an overview of how many images are similar and in what similarity range. To view the result of a range click on a bar. None solid bars will display 1000 pairs only. The red vertival line marks the set threshold.
Investigate Images
Hover over an element or button to obtain more information about its function. Click on a image to enlarge all by 5%. Use the right mouse button to shrink the images by 5%. Hold the left mouse button down and move the pointer to pan the images. The images size can be influenced via the window size or by moving the splitter between the center and right image.
For the quick assessment of differences, several difference illustrations (center image) are available. The flicker method displays alternating the left and right image, which lets differences flicker and easy to spot. The current image will be indicated with a little color bar where red indicates the larger one. Also in red are words describing the differeneces.
File names can be very long, that's why the information fields enlarge with a click into them. A mouse click outside an information fields will restore its normal size. Text can be selected and copied to the clipboard.
The copy buttons will open a new dialog with lots of choices on how to copy and delete after copy.
There are also buttons to open an explorer window for the left or right folder.
The color of the information field is changed on directory change to make a change easy to notice.
The number of files in that folder is displayed in brackets behind the creation time.
The background color of the center area indicates the position of the images in the folder structure. Gray is same folder, yellow different folders, lightblue is if one is in a child folder and blue if both images in folders with the same parent.
Color Pattern Search
Place a few color patterns on the canvas and hit OK to find images which closely match to these color patterns and not taking canvas color areas into account.
Note: The similarity is actually a ranking for this type of search. Also note: White contains all other colors. This means different colors are mathematically similar independent of your perception.
Here are a few samples, which show how easy it is to find pictures in a database. To the left is what was painted in the box above and the right most images is the top ranking result.
Ignore List
In case archives are often compared to each other, one can easily find the same picture pairs again and again. This is of course not wanted and thats why one can add those pairs to an ignore list. Such picture pairs, ones add, are no longer presented as results.
Via Ignore list maintenance all picture pairs are displayed in the result list. Excluded are those pairs where not both parts exists. In the list entries can be deleted or be added to the now empty background list (not visible). On dialog change one will be asked which of the lists should remain.
More in A-Z