In this example the gray horizonal stripe is not percived as uniform. It appears brigher next to the black rectangles and darker next to the white rectanges. Our result compensates for this bias and produces an image with the counter variation which is percived as having a more uniform gray stripe.
Chevreul's illusion
Input
The columns in the input image have the same gray-level value yet appear as if they are darker on their right side and brighter on their left.
Our correction reduces this effect. This illusion is seen best when maintinaing a fixed gaze at the interface between two columns.
Mach Bands
Input
In this well-known example there are two false lines; a dark one at the left side of the gradient and a bright one at its right.
These lines are reduced in our result.
Simultaneous Contrast
Input
While the two circles have the same gray level value in the input image, they do not appear identical due to their different backgound color.
Our method makes it easier to detect the brightness constancy in this example.
Bars
Input
Here our method reduces the false brightening around the Google Scholar bars.
Youtube page
Input
This examples shows a screen capture from YouTube where halos can be seen around the video's frame.
These halos result from the high contrast edge between the frame and the white background. They are reduced in our compensated output.
SIGGRAPH webpage
Input
Another example of an HTML page in which false variations appear. Notice how the lower edge of the blue rectangle appears brighter next to the black space pixels and darker next to the brighter earth pixels.
In our compensated result this false variation diminishes.
Computer Graphics Forum webpage
Input
In this HTML example the transition bewteen the tab menu at the top of the page and the white background produces an unpleasant percieved edge. Bright halos appears around the logo and the main banner as well.
In our compensated result this false variations is reduced dramatically.
Underground logo
Input
A bright halo is seen around the red circle as well as at the the interface with the blue rectangle (mostly on the red side).
Our result reduces this effect and is possibly closer to the logo designer's intent.
Coca-Cola logo
Input
Bright halos around letters are reduced by our compensation.
Ebay logo
Input
Bright halos around letters are reduced by our compensation.
Yin Yang
Input
False halos along the black and white edges are observed in the input and corrected by our method.
Architecture plans
Input
False brightening is percived around the black walls (especially at the grayed rooms). These biases are reduced by our method.
Maps
Input
In this example, false halos appear around the tube lines, especially around the black ones. Our laterally-compensated image offers a softer appearance.
Tetris
Input
The halos bewteen two adjacent pieces are reduced in the compensated image.
Packman
Input
Another example of a computer game where halos precived across the screen.
Timetable
Input
The time-table cells exhibit disturbing variations due to their different brightness levels. Our compensated table has a sofer appearence.
Pie-chart
Input
An example of a pie-chart where false halos along the cross-sections are obsereved in the input and reduced by our method.
Silhouette Image
Input
The strong halos along the man's silhouette are greatly reduced by our method.
Garfield
Input
A bright halo appears around Garfield's outline.
In our compensated image the edges are less harsh and the background appears more uniform.
Simpsons
Input
The halos around the figures are reduced by our method.
Simpsons
There many cases where halos are seen in the input clip and are reduced in the compensated one. The video pauses and shows some of the corrections.
Below we show these clips side by side.
Southpark
Input VS. Output Simpsons
Input
Compensated
Video clips shown side by side. The processed video has a softer appearence which makes it easier to watch.
The "Play" button applies to both clips at once.
Input VS. Output Southpark
Input
Compensated
X-ray Hand
Input
Left image shows an X-ray image and the 1D plot shows the intensity levels of the pixels along a horizontal scanline inside the orange rectangle.
While the change from the soft tissue to the bone is monotoic in the plot, an oscillation is percived when watching the image.
As we noted in the paper, this issue raises concerns in the medical imaging community. Our result gives a better depiction of the data being visualized by reducing the percived oscillations.
X-ray Arm
Input
Another example case as the above.
X-ray 1
Input
In this example the overlap between two types of maxillary molar roots (orange rectangle) produces a thin black line in the transition.
This line does not exist and the transition from one root to the other is monotonic.
X-ray 2
Input
The white arrow is pointing to the interface between the dentin and the gutta-percha (sealer). The inconsistency between the regions might cause an incorrect diagnosis of vertical root fracture.*
In the output image the effect is reduced.
X-ray 3
Input
In this example, the white arrow is pointing to the overlap between two types of roots. The transition between these two regions produces bright and dark thin lines.
A clinican might identify these lines as the periodontal ligament.*
In our compensated image these thin lines does not exist.
X-ray 4
Input
The overlap of the opacified arteries creates an effect of inconsistancy between the regions that might be misinterpreted as an intraluminal filling defect.**
X-ray 5
Input
In this example we can see a lower molar after obturating. The arrows are pointing to the area which might be interpreted as a void that warrent retreatment.
However, the thin black line is an illusion caused at the edges of the filled molar canals.*
X-ray 6
Input
Similarly to X-ray 2 image, the transition between the dentin and thegutta-percha (sealer), in the orange rectangles, produces false halos.
Our result gives a better depiction of the data by reducing the percived effect.
Cameraman
Input
The interface bewteen the bright sky and the dark shoulder produces a thin bright halo. While this is how we normally percive natural scenes, our result offers a softer appearence which may be preferable in some situations.
We should note that the halos percived in the input image do not exist in the image (its gray-scale values change monotonically along the strong edges such as the shoulder).
Woman
Input
Similarly to the example above, the strong halos along the woman's silhouette are reduced by our method.
Chair
Input
Another example where thin halo is percived around the chair.
Father and Son
Input
Here some halos are percived around the faces (cheeks) at the center of the image due to the very bright background.
Obama
Input
The halo around Obama's head does not exist in the image pixel values. This effect is greatly reduced in our result.
(although we could see how some will prefer to preserve this holy glow..)
In this example we show how the details preservation mechanism contributes to the overall appearance while not undermining our lateral correction.
Lamp
Input
Enhanced edges are percived between the left side of the hand and the sky. This effect is eliminated in our compensated image.
Horses on the moon
Input
The halos around the horses are reduced significantly in the compensated image.
Sunset
Input
The strong halos around the giraffes and the tree, caused by the high-contrast between the subjects and the background, are significaly reduced in the compensated image.
* Image taken from "Effect of Scenario and Experience on Interpretation of Mach Bands". Nielsen C. J. Journal of Endodontics 27, 11 (2001), 687-691.
** Image taken from "Angiographic artifacts that simulate arterial pathology in acute trauma". Rose S, Moore E., Am J Emerg Med 1990;8:109-17.
