Access to eBooks is limited to institutions that have purchased or currently subscribe to the SPIE eBooks program. eBooks are not available via an individual subscription. SPIE books (print and digital) may be purchased individually on SPIE.Org.

Contact your librarian to recommend SPIE eBooks for your organization.
Chapter 12:
Image Simulations
Abstract

12.1 Putting It All Together: Image Simulations from the Imaging Chain Model

Now that we have learned about the mathematical models used to describe the key aspects of the imaging chain, we can put them together to simulate images that would be produced from various camera designs. Figure 12.1 illustrates the process flow for generating image simulations. Note that the process essentially walks through the imaging chain up to and including the image processing, after which the image simulations are displayed and interpreted to understand and quantify the image quality.

12.1.1 Input scene

Perhaps the most difficult aspect of generating image simulations occurs at the very beginning of the process, when input scene data must be found that properly characterizes the scene attributes. Ideally we would like the input scene to have infinite spatial resolution and infinite SNR with all of the relevant material properties, such as spectral, reflection, and scattering, for every point in the scene. Synthetic scenes generated from physics-based models, such as Rochester Institute of Technology's (RIT) Digital Imaging and Remote Sensing Image Generation (DIRSIG) model, will produce input scenes with high spatial and spectral resolution, but the generation of this data is very time consuming, so the variety of scenes is limited. It is also very difficult to generate enough fine details in the scenes to prevent them from having a cartoon look to them.

Another option is to use the scene information from an existing image. This option is much simpler than a computer-generated scene but has limitations. In order to properly simulate the imaging chain effects such as blurring and sampling, the input image must have significantly better spatial resolution than the desired output image. A general rule of thumb is that the input image should be at least 4X; better resolution then the desired resolution of the image simulation. The input image should also have a very high SNR so that the noise present in the input image does not impact the desired image simulation. The input image should match the spectral bandpass of the desired simulation. (The desired spectral bandpass can also be created if the input image is a hyperspectral image with high spectral resolution.) Also, the radiometry can be modeled for different sun angle conditions, but the shadows will be fixed in the simulation to the sun angle conditions of the input image. Finally, the input image will have all of the physical properties of the imaging chain from the camera that collected it, which must be accounted for in the imaging chain model; thus, it is desirable to use images from cameras that have been well characterized.

Online access to SPIE eBooks is limited to subscribing institutions.
CHAPTER 12
16 PAGES


SHARE
Back to Top