Demystifying 3D Product Photography in eCommerce
With the proliferation of 3D scanners and 3D scanning technologies, such as structure from motion, there’s more talk about 3D imaging for eCommerce application. While “3D imaging” is a term fairly abused in eCommerce, I’d like to clarify the true meaning of some of the terms that are replaced by “3D imaging.” The goal is to educate about different eCommerce imaging technologies. When relevant, I’ll reference imaging technologies that our team have developed at Iconasys.
Here are the ten most common eCommerce imaging techniques:
1. Still Product Photography. This most often refers to a dense matrix consisting of many pixels, arranged in rows and columns (hence the name 2D, one for row and one for column) of different colors. This is the most common type of imaging technology used in eCommerce and it is accessible in both print and digital media. 2D images are captured by standard cameras. With processing, some image formats allow for an alpha channel, which is a weight that tells other programs how to combine the image with a secondary image, most often a background image. Some sample file formats are JPG, TIF, PNG, BMP and others.
2. Movies and animation. These are 2D images displayed in pre-defined sequences, most often in time, but sometimes they may also be sequenced in a 2D space (think of a 2D matrix of images). Movies and animations are created by capturing or composing a sequence of 2D images. This is the second most popular visualization technology in eCommerce. The order of 2D images cannot be changed, but the speed and direction with which one flips through the sequence of 2D images can be controlled in certain situations. When the movie playback is interactive, the movie control creates the impression that the user controls the turn of the object. These movies can be embedded in 3D environments to give the impression that the movie is a 3D object. In my opinion, calling a movie “3D imaging” is misleading, regardless of how the movie playback is done. Example of Apparel Movie extracted to make interactive view.
At Iconasys we developed the 360 Product View Creator Software that allows users to create interactive movies – a user simply drags the mouse in different directions and different frames of the movie are shown. Further, our software allows users to add hotspots that link to other websites, 2D images or other movies. The interactive movies can be embedded into websites that allow Java scripting and they can be hosted on a proprietary server or Iconasys’ Streaming Server.
Some movie and animation file formats are MPEG, MP4 and animated GIF. There are currently no standard formats for interactive movies. At Iconasys we developed our proprietary viewers that use MP4 movies or collection of JPG/PNG images with playback controls written in Java Script.
3. 360 Product Photography or 360 Product Views. These are movies taken by a static camera of objects that turn on a rotating turntable or by a camera that moves in a circle around a static object. At Iconasys we provide 360 Product Photography Turntables and 360 Product Photography Software that automate 360 image capture in a turn, stop, snap workflow. The ability to image a product when the turntable is stopped is critical for achieving the maximum quality per-frame image and to minimize blur and noise. Typical file formats are HTML5, Interactive MP4, Animated GIF.
4. Multi-Row 360 Product Photography. Also referred to as 3D Product Photography, these are multiple rows of 360 images stacked on top of each other, basically a matrix of images. Moving the mouse left to right flips images along the row and moving the mouse up and down flips images along the column. 360 multi-row images are captured either by multiple cameras sitting along a bent arm, each doing a 360 view or one camera doing multiple 360 views as the camera is moved from the side to the top of the object.
5. Panoramic images.
These are 2D images projected on a 3D sphere and viewed using a panoramic viewer. These images are very effective in showing the inside of rooms or cars and they are most often used in real-estate and car sales. The camera inside the 3D sphere is rotated at the origin to see different parts of the sphere and hence of the panoramic image. The panoramic image is a 2D image, with the width usually twice the height. This is because there are 360 degrees of horizontal views and 180 degrees of up-down views, hence the 2:1 ratio. Image capture is often done with a panoramic camera or with a camera sitting on a tripod and capturing multiple images as the camera turns horizontally and vertically. Images are then stitched together into a panoramic image.
At Iconasys we developed a panoramic server that allows users to upload panoramic images and then embed the panoramic viewer into any website that supports Java scripting. The pano-server we developed also allows users to embed links to other panoramas, movies, websites or images.
6. 2.5D images or depth images. These are 2D images where each pixel provides a depth information, in addition to color. 2.5D images are most often associated with geo-spatial imaging: a LIDAR in tandem with a 2D camera captures the depth (distance from sensor to ground) and the 2D image. Post processing registers the 2D image with the depth image and together they can be used to generate orthographic images. The reason that these images are called 2.5D and not 3D can be explained by thinking of how a LIDAR sensor, on an airplane, scans a building. It usually only sees the height of the building and not its side. More details on 2.5D are available on Wikipedia.
The latest mobile phones have the ability to capture depth images. When multiple depth images, from different angles of the same object, are merged together, 3D point clouds and 3D surfaces can be generated. Depth images are starting to play a bigger role in eCommerce as they can provide size measurements and their role will only increase in the future.
7. Stereoscopic images. These are sets of 2D images that vary slightly and are shown to the left and right eye separately in order to create the sensation of depth. To separate the two images, stereoscopic images often require the use of special glasses. A few years ago, a lot of 3D TVs were nothing else but stereoscopic TVs. For some users, using stereoscopic images can throw the brain for a loop due to the lack of ability to control focus. Imagine that you are looking along a hallway and the stereoscopic images are focused on the far end of the hallway. If you were looking at a true hallway, your eyes would be able to change focus from the far end to the near end. The lack of ability to change focus in stereoscopic images is one reason that some users get headaches and stereoscopic images are not recommended for young users whose vision system is still under development. Stereoscopic images are not very common in eCommerce.
8. 3D point clouds. These are collections of points, where each point has a 3D location (i.e. an (x,y,z) coordinate) and may have, but not necessarily, an RGB color. Point clouds tend to be sparse (i.e. points are not next to each other on a regular grid, but rather sparsely distributed) and the order in which point clouds are listed does not change the visualization of the data.
The latest mobile phones provide sensors that capture depth images and multiple depth images can be registered together to generate complex point clouds. Direct point cloud visualization may be challenging for eCommerce but point clouds may be a stepping stone to other products that may find more usage for eCommerce, such as 3D surfaces.
9. 3D surfaces. These are 3D point clouds where points are connected together to create surfaces in 3D. Most often, points are connected via triangles and within a triangle the texture is a 2D image. 3D surfaces are usually generated out of 3D point clouds and they are the holy grail of 3D imaging in eCommerce. The future is 3D surfaces, but at the moment there are still many challenges left in capturing good 3D surfaces, two of which are:
a. 3D surfaces of shiny, transparent or dark objects are hard to capture.
b. 3D spatial and depth resolution can be a challenge, especially in objects with holes or high frequency content, such as hair.
On the playback end of viewing 3D surfaces, there have been studies that have shown most average users can easily get lost manipulating a 3D surface and frustration can easily set in. Interactive 360 images and movies may provide less freedom, but more control in guiding an average eCommerce user.
In time, all these challenges will be overcome and 3D surfaces are the future of eCommerce imaging.
10. 3D Volumes. These are usually densely gridded 3D point clouds. Each 3D point (voxel) has a neighbor. 3D volumes are most often associated with medical images, such as CT scans or industrial applications and they are unlikely to see a big use in eCommerce.
In conclusion, we presented ten of the most common eCommerce imaging methods and discussed their properties from a high-level point of view. The future of eCommerce is definitely 3D surfaces. Currently, a very effective method for eCommerce imaging is interactive playback of movies and 360 views.
About the Author: Darian Mursean is the CTO of Iconasys. Darian has 20 years of relevant technology and domain knowledge in multi-dimensional signal processing, software and hardware engineering. Previous work experience includes: Walker, Richer and Quinn (WRQ), Hewlet Packerd (HP), Solers, Geo-Spatial Technologies Inc. (GSTI), Science Applications International Corporation (SAIC), Van Gogh Imaging (VGI), Leidos, Geospatial Intelligence and Analysis (GIA) and Digital Multi-Media Design (DMMD). Highlights include: principal investigator (PI) for several light detection and ranging (LiDAR)-based research and development programs focused on streaming LiDAR and processing Geiger mode LiDAR; developer of XRay image processing algorithms for a Food and Drug Administration (FDA) approved medical software suite; and holder of several image processing patents, including a patent for the development of a “Process and System for Three-Dimensional Urban Modeling” (Patent No. 7,752,483 B1, July 2010), which was developed under Defense Advanced Research Projects Agency’s (DARPA’s) Exploitation of 3D Data (E3D/3DF) and UrbanScape programs.