Significant processing is necessary in order to transform a RAW image coming from a sensor into a high-quality image. The image sensor produces linear data up until the point of saturation. After that the Image Signal Processing (ISP) performs many of these steps:

  • Lens shading correction (LSC) is applied to improve the uniformity of the illumination and color.
  • Demosaicing converts a single channel Bayer RAW image into a three-color image.
  • Chromatic aberration can be partially corrected by resampling the different color channels.
  • Gamma encoding applies a tonal response curve to the linear sensor data. S-curves or shoulders may be added to prevent highlights from burning out.
  • White-balance & Color correction applies color correction matrix (CCM) transforms to adjust the colors to fit a particular output color space (such as sRGB). This process may include making an estimate of what sort of illuminant the scene contains so that different corrections can be performed for different lighting conditions.
  • Sharpening is applied to increase the appeal of the image, too much can lead to fringing artifacts.
  • Noise reduction is applied more when the sensor is set to a higher sensitivity setting and applied to areas of the image away from the edges. This processing should take care to avoid removing legitimate textures in the scene, particularly within skin tones.
  • Local Tone Mapping adjusts the brightness of different regions of the image or combines different exposures together in order to increase the local contrast within disparate regions of an HDR scene.

See Also