Image Matching

LTU's image matching technology is able to automatically and very efficeintly detect that two images are clones, or find clones of an image within a database of images. Of course, these two images can be frames extracted from a video file.

The image matching ability of LTU Engine is robust to image transformations, detecting not only the exact same image but also image clones. Depending on their needs, our clients can choose between a crop robust local matching DNA or an extremely fast global matching technology that will detect fewer clones, but much more rapidly.

We call two images clones if they are both the result of editing, redacting, photoshopping, resizing, reencoding, the same original image...

In a more mathematical formulation, the array of pixels in an image is the results of the convolution of a signal function and a rendering function. We call two images "clones" if they have the same signal function, but different rendering functions.

Image(x, y) = signal(x, y) o rendering(x, y)

Local Matching

If two images only have a part in common, then our local matching technology can detect that these images as clones. Following are various examples of what can be detected. Cropped images - One image can be a "cropped" version of the other, with additional editions like color changing, reenconding, etc.. Composite image - One image can be made of several other images. Our local matching technology will detect all original images it was made of. Photometric and statistic noise - in addition to the above geomoetric transformations, our local matching technology is highly robust to: Hue, contrast, luminance, palette changes, posterization,... Any type of gaussian, hessian or other geometric noise Resizing, reencoding

 Global Matching