Introduction to Programmable DSP Platform for DSC

1
Introduction to Programmable DSP Platform for DSC
Color Imaging

• B88901136 ???

2
Digital Photography Market Overview

• Major classification
• Soft Display (PC camera)
• Basic point-and-shoot (low-end consumer)
• Photo-quality point-and-shoot (high-end consumer)
• Professional

3
DSC Market Size and Growth
4
DSC System
5
Image Acquisition

• Nowadays,most cameras use CCDs more than CMOS.
• CFA pattern is mostly common used with an RGB
  Bayer pattern.(2x2 cell elements that are tiled
  across the entire CCD-array)

CCD
6
Image Pipeline
7
Image Pipeline (1)

• Black Campsubtracting an offset from each pixel
  value,but clamping to zero to avoid a negative
  result.
• Lens Distortion Compensationreduce the
  brightness between the center of the image to the
  border of the image.
• Fault Pixel Detectionthe image values at the
  faulty pixel locations are computed by an
  interpolation technique.

8
Image Pipeline (2)

• White Balance to compensate for the different
  sensitivity of each color channel,the gain of the
  red,green,and blue channels is equalized.
• CFA interpolationthe missing pixels are
  reconstructed by interpolating the neighboring
  pixels.
• Gamma Correctioncompensate for the difference
  between CCD sensor and monitor display.

9
Image Pipeline (3)

• Color Space Conversionat a pixel
  location,transform RGB to YCbCr(for JPEG
  compression).
• Edge Enhancement compute and add the edge
  magnitude in the Y channel at each pixel.
• False Color Suppressionsuppressing the color
  components,Cb and Cr,at the edges reduces come
  rainbow-like artifacts.

10
Image Pipeline (4)

• Autofocuscan operate in a feedback loop to
  sharply capture the photos automatically.
• Autoexposuresensing the average scene brightness
  and appropriately adjusting the CCD exposure time
  or gain.
• Image Compressionmost DSCs use JPEG
  compression(computation of DCT Huffman
  encoding).Future will turn to JPEG2000.

11
TMS320C54x DSC Implementation

• The image pipeline was implemented on a
  C5409DSP.All pipeline operation can be
  implemented on TMS320C540x and bit stream is
  written out to the SDRAM.

12
Efficiency of TMS320C54x

• The entire image pipeline,including JPEG,takes
  about 150 cycles/pixel.With 100MHZ
  computation,1M-pixel image can be processed in
  1.5sec.

13
Memory Requirements

• The program and data memory requirements to
  process the image pipeline and compress the image
  with JPEG standard is shown below.This allows the
  complete image pipeline software to reside
  on-chip.

14
Reference Data

• Color filter options
• Color processing
• Color interpolation (demozaicing)
• White Balance
• Color correction

15
Color Filter options

• Use three image sensors and a beam splitter.
• Use time-switched color filter(each color can
  have different exposure time).
• Use CFA or mosaic deposited on top of the pixel
  array,each pixel outputs only one color
  component,e.g.,R,G,B.

16
Color Filter Arrays

• Lowest cost option
• Lower spatial resolution,optical loss due to
  filters
• Processing need to reconstruct the missing color
  pixel

17
Color Processing

• (i)reconstruct missing pixel color
  component(ii)produce color that is close to what
  the eye would receive

18
Color Processing Steps

• 1.color interpolation (digital domain)2.white
  balancing (analog domain) 3.color correction
  (digital domain) 4.gamma correction (digital
  domain) 5.color conversion (digital domain)
• It is computationally very demanding (about 70
  of processing in a digital camera is related to
  color)

19
Color interpolation

• Used to reconstruct the missing pixel color
  components(when CFA is used).
• Algorithms1.nearest neighbor replication2.bilin
  ear interpolation3.2D-filtering
• Adaptive algorithms-gt(i)change method at
  edge(ii)more edge detection

20
Color interpolation Examples
21
White Balancing

• Different light sources (illuminants) have
  different power spectral densities(psd).Captured
  images must also be processed so that a white
  background looks white.

22
Gamma Correction

• Display device is not linear to input.
• Must prewarp the image sensoroutput.
• Typically implemented using 3 lookup tables , one
  for each color component.

23
Color Space conversion

• Transform RGB to YCbCr,or to YUV using 3x3
  matrix.
• Use least squares method to find D.