Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from P-T

Spectral Model

image processing color characteristics

Definition: Single-sensor solutions may employ a spectral model in the processing pipeline in order to take advantage of both the spatial and spectral properties of the available samples and thus eliminate color shifts and artifacts in output, single-sensor captured images.

The fundamental rationale behind the introduction of the spectral model is that a typical natural image exhibits significant spectral correlation among its Red-Green-Blue (RGB) color planes, as well as significant spatial correlation among its spatially neighboring pixels. Since natural color images do not have large areas exhibiting uniform image characteristics or small variations in pixels’ values, modeling constraints are applied to spatial locations neighboring the pixel location selected for processing. Smoothly changing spectral image characteristics are enforced by averaging the obtained spectral quantities in the considered spatial neighborhood.

The currently used spectral models are based on the assumption of color hue or intensity constancy in localized image regions. However, these modeling assumptions often break down in high-frequency image regions. Recently proposed designs are able to overcome the limitations of the basic spectral modeling approaches, provide more degrees of freedom in modeling the spectral characteristics of the captured image, and are able to enforce the modeling assumption in both flat and high-frequency areas.

Since the human visual system is more sensitive to luminance information which is composed primarily of green light, most spectral models are based on a two-channel information (RG or BG) base. Such an approach is sufficient in color filter array (CFA) image zooming applications, however, it is of limited value in both the demosaicking and the postprocessing/enhancement phase of the single-sensor imaging pipeline. To reduce processing errors, a vector model should be employed in order to utilize the complete available spectral information. The vector model is a generalized spectral model which can be used to support processing of multi-dimensional data sets such as the captured image data. Through the refined tuning of its behavior, the vector spectral model controls the influence of both the directional and the magnitude characteristics of the neighboring pixels utilized during processing. New single-sensor solutions, with different design characteristics and performance, can be obtained by modifying the way the vector model is operating.

A generalized vector model can be made flexible enough to support any camera image processing operations such as demosaicking, demosaicked image postprocessing and CFA zooming. In addition, it can be used in applications such as computer graphics and color image processing to support denoising, image analysis, and color constancy algorithms. Thus, it constitutes a unique and powerful solution for the image processing pipeline.

Speculum Humanae Salvationis [next] [back] Spectral Imaging

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or