Align Image

The Align Image module provides a way to align two images. Similar to the Visual Anchor and Stabilize module the Align_Image module will look for characteristic properties between two images in an effort to determine how they overlap.

Note that if too much of the target image is not included in the image to align then the system will produce erratic results as any detected alignment between the two images will be random.

This module will determine the translation (X,Y) movement, the Z plane rotation or orientation and the scale changes between the two images. Note that in order to associate the placement of one image over the next you will need some texture within the image that is the same in both in order to determine how to overlay the image.

The module was developed to allow for comparison between two images that may be somewhat misaligned such as in a security setting to determine abandoned or removed objects. It is also very useful in document comparison such as comparing differences between scanned or otherwise photographed documents. Both automated and manual modes of alignment are supported.

Interface

Instructions

1. Target Image - Select the "Target Image" in the Image Source dropdown. The target image is your model image that typically does not change. The source image will be stretched, rotated, etc. in order to make it as close as possible to the Target Image.

2. Align Image - Select the image you want to align to another image. Typically this will be the current video image from your webcam or come from a loaded image (see Load Module) for image loading.

3. Options - Select which mode you would like to use for image Alignment. The Automatic mode seeks to find the correlation between the two images without needing any feedback from you. There are cases that may occur when you will need to adjust the values manually. In those cases you can select the Manual radio button which will enable all the value editing interfaces in the Manual Values section.

In automatic mode the "Text & Line Images" selection will work better with images that are mainly black and white and contain lines, drawings. The "Color Images" works better on images that are of real scenes and contain texture/gradients within the image.

Note that in order for the automatic mode to work correctly the majority of the image content needs to be shared between the two images.

4. X, Y Translation - These are the values associated with the horizontal and vertical movement of the Source image to align it with the Target image. Modifying this value will move the Source image in relation to the Target image in efforts to align them better. Note that these values are in pixels.

5. X,Y Scale - These are the values associated with the size change from the Source to the Target image. For example, an X or Y scale of 2 means that the Source image is enlarged to twice its size in order to align with the target image. Similarly, a value of 0.5 means the Source is reduced to half its size in order to align with the target image. One can interpret this as meaning how close or far are you from the Target image in terms of depth.

Typically images scale in the same amount for X and Y. But sometimes due to a non- perpendicular photographic plane this may not be true. If you select the Automatic mode or press the Refine button both X and Y will be optimized separately just in case. While manually editing you may find it easier for them to have the same value which can be enforce by checking the "Scale X&Y together".

6. X, Y Rotation - These are known as 'out of plane' rotation in that they will spin and flip the image (X and Y Rotation) such that the image appears to exist into and out of the computer screen (ie. the Z Plane). Note that the automatic mode does NOT currently change these values.

7.Z Rotation - This is also known as orientation and fixes situations where the Source and Target image are upside down with respect to each other. This parameter will twist the Source image until it better aligns with the Target image. If you are more comfortable in working with Radians as apposed to degrees you can select the Radian radio button to switch between Degrees and Radians.

8. Refine - The refine button will execute an automated alignment just once on the image. This is helpful when using the manual mode of image alignment such that when you get things close you can try the Refine button in order to tweak the final values. Note that the alignment should be close in order for the Refine button to work. If the images are far apart in terms of alignment then the Refine button may not find a match or make things even worse.

9. Pixel, Bilinear Resample - Select how precise the Source image should be transformed into the Target image. Pixel Resample refers to a nearest neighbor approach where a pixel from the Source image is transformed and placed into the nearest pixel of the transformed image. This creates a jagged effect but provides a quicker transform process. The Bilinear will interpolate the pixels position and smooth out the final image to avoid jaggies and create a more natural final transform but at the expense of reduced performance.

10. Perspective - While Rotating in the X and Y plane (spin and flip) you can specify the amount of perspective to incorporate into the transform. The amount of perspective will change how much the image shrinks at points far away in depth and how much the image enlarges when closer in depth. A high perspective number causes things to seem very fat while a smaller number will cause more and more warping of the image at higher X or Y rotations. The amount of Perspective loosely correlates with the amount of warping seen in the image which is interpreted as depth by us.

Example

Source Target

Aligned Image

Download the zip file containing the above two images and the robofile used to generate the aligned image.