Approximate Calculation in Image Processing

(Master of Information Science) Lecture on IoT and Edge Computing

• Quite skippable
• For example, the empty areas of an image often have pixels that are similar in color.
• If we can reuse the results of calculations we have done before, we can skip some work (Approximation Calculation).
  • Challenge
    • How can we inform the CPU about the opportunities for approximation?
      • “Specify the address where the software instructions are stored in advance.”
      • Give the recently calculated/used data some “freshness” like quality.
      • Save it in the Data Store Table (DST) for reuse.
• Image compression algorithm#image processing
  • Heavy processing areas:
  • Conversion from RGB to YCbCr
  • Discrete Cosine Transform (DCT)
• Using DST for Approximation Calculation in image compression
  • Experimental result: By selecting the right combination of parameters, we can reduce computational complexity, among other things (compared to a regular CPU).