In the final rendering, the researchers have used the method published in Applied Optics where you allow for a full calculation of the refractions of light and the fact that the glass also absorbs a small portion of the light. Photo: DTU Compute og Applied Optics, Jeppe Revall Frisvad

Researchers create images with mathematics

Image analysis Innovation and product development Computer calculations

Researchers in the field of image analysis have managed to create images of glass objects with photographic precision—without actually photographing them.

The boundary between reality and a virtual world has been further blurred after a team of researchers have developed a new method that can reproduce glass objects with photographic precision—without photographing them—it should be noted.

The images are digitally produced using computers which are fed with mathematical models, algorithms, and data from scans of the objects.

The digital reproductions of the real glass objects are so realistic that the researchers had their findings published in the scientific journal Applied Optics in autumn 2017 and enjoyed a good deal of media coverage—especially in the United States.

“The eye is easily fooled when we look at complex optical materials like glass, which is probably the most challenging material to digitize. We have therefore developed a method which makes it possible to determine the degree to which digital glass corresponds to real glass,” says Professor Emeritus Knut Conradsen, who was part of the team comprising 13 researchers from DTU Compute and one researcher from DTU Physics.

“Clear glass is almost invisible. It is the light and its refraction through the material that enables people to see it. At the same time, glass reflects light in many different ways. It is therefore considered something of a feat to achieve such realistic rendering of glass objects,” says Knut Conradsen.

Virtual glass objects

The researchers had set out to make a digital copy of a real arrangement of three glass objects—a jug, a ball, and a platter with a lid—on a chequered background illuminated with LED lights.

Photo: Jeppe Revall Friisvad, DTU Compute and Applied Optics
The top row of images are computer-generated renderings of three glass objects. The three pictures below are the actual photos of the same objects. Photo: DTU Compute and Applied Optics.

Photo: DTU Compute and Applied Optic
Subsequently, the researchers merged the digital glass objects with a photo of a coffee table. It is impossible to see that the glass jug, dish, and the ball have not always stood on the coffee table. Photo: DTU Compute and Applied Optics.

Light calculations

The three glass objects were then CT scanned so that the researchers could obtain information on the design of the objects, which they subsequently translated into mathematical models.

The researchers then combined this information with models of how light rays move through different objects. Finally, the many data were compiled into the realistic images. Using the original photos of the glass objects as a reference, the researchers were able to compare the two versions pixel-to-pixel.

“The similarity is remarkable. If you look closely, you can spot a tiny difference in the light dispersal in the digital version. That said, we have shown that we can recreate glass objects digitally—not just because we have a powerful computer, but also because we have an understanding of what is taking place.  This enables us to make mathematical models of how reality is behaving,” says Knut Conradsen.

Images reproduce reality

The method of creating virtual glass objects can be useful in film production and virtual reality experiences where the requirements for absolute lifelike reproduction are not as demanding. More relevant applications are to be found in prototyping, design improvements of products, and quality assurance.

But the DTU researchers have proved a wider point: With such advanced analytics capabilities it is reasonable to assume that the renderings created using data from e.g. X-ray synchrotons and which visualize enzymes, nerve pathways, or fibres in a wind turbine blade, actually show us a world as it would look if we were able to photograph it.

Firstly, each object was photographed by two cameras. Secondly, the researchers used a robot arm to record the exact location of the two cameras in order to obtain data on the spatial dimension of the setting, which is crucial in order to digitally recreate the arrangement.

 Photo: Jeppe Revall Friisvad  

 

The three glass elephants are all computer-generated renderings and there is no photo of a real glass elephant.

The renderings move from a simple rendition of glass to increasingly advanced renderings of the glass elephant.

Notice how the reflection of the person in the background is changed in the last two renderings, where the researchers have added calculations of the light’s refraction through the material.

The rendering illustrates how we can be fooled into believing that something is glass—even if it distorts the light only slightly. 

 

Here, the rendering contain some of the calculations of the light’s refraction through the glass. However, if we study the details, we can see certain shortcomings, as the image lacks absorption as well as many refractions and internal light reflections.

 

 

 

 

In this final rendering, the researchers have used the method published in Applied Optics where you allow for a full calculation of the refractions of light and the fact that the glass also absorbs a small portion of the light.


Photos: Jeppe Revall Friisvad.