markus holzer: accelerating form based image preprocessing with digital hardware
DESCRIPTION
Presentation at the 10th Heidelberg Innovation Forum, http://www.heidelbergerinnovationsforum.de/TRANSCRIPT
10th Heidelberg Innovation Forum, Studio Villa Bosch 12th April 2011
Markus Holzer, Thomas Greiner
Pforzheim University
Center for Applied Research - MERSES
Accelerating Form Based Image Preprocessing with Digital Hardware
Outline
• Introduction to form based image processing with Morphological Operations
• Novel principle (OSLCR) of efficient digital hardware realization of Morphological Operations
• Architectural performance and hardware complexity of OSLCR
• Requirements concerning the transfer business
Non-linear neighborhood
operations Discrete 2D Image
processing
What are Morphological Operations?
Introduction to Form Based ImageProcessing with Morphological Operations
Application fields
• Image filtering / enhancement• Noise reduction, object contour smoothing
• Object segmentation• Content based image/video coding/compression• Pre-processing for computer vision
• Object analysis and measurement • Object summaries related to e.g. form attributes,
texture, orientation (e.g. for granulometry)• Exploration object topology (e.g. for OCR)
Major advantages• Can remove artifacts without smearing significant object edges• Efficiently implementable (especially for binary images)
Introduction to Form Based ImageProcessing with Morphological Operations
Description of Morphological Operations
• Compositions and repetitions of two basic operations: erosion and dilation– 2D gray level input image G– 2D binary / gray level signal: flat / non-flat
structuring element (SE) defines the decisive form signature
• The Input Image is probed with SE– SE is shifted over the entire input
imagepixel-wise in raster scan mode
– For each shift step: • The minimum / maximum in the
scope of SE must be found
• The pixel of output image G congruent to the actual reference point of SE is set to this minimum / maximum value
reference pixelflat structuring element
Example
G
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
Example
GG
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
Example
GG
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
Example
GG
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
Example
GG
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
Example
GG
Erosion with flat SE:8 bit gray level input image
13 ×13 diamond shaped SE
• Observation: between subsequent line wise shift steps
Example
G
Erosion with flat SE:8 bit gray level input image
Example
→ pixel overlay → redundant comparisons→ for large-area SE direct implementation is inefficient
– direct implementation of 13×13 diamond shaped SE→ 84 comparisons per shift step
G
Erosion with flat SE:8 bit gray level input image
OSLCR Architecture Principle
OSLCR: Orthogonal Shift Level Comparison Reuse •SE shape independent digital hardware design approach•Principle: Comparisons along discrete shift levels (DSL)
are reused (orthogonal to the shift direction) while the SE is displaced line-wise over the image
OSLCR Architecture Principle
OSLCR: Orthogonal Shift Level Comparison Reuse •SE shape independent digital hardware design approach•Principle: Comparisons along discrete shift levels (DSL)
are reused (orthogonal to the shift direction) while the SE is displaced line-wise over the image
OSLCR Architecture Principle
OSLCR: Orthogonal Shift Level Comparison Reuse •SE shape independent digital hardware design approach•Principle: Comparisons along discrete shift levels (DSL)
are reused (orthogonal to the shift direction) while the SE is displaced line-wise over the image
OSLCR Architecture Principle
OSLCR: Orthogonal Shift Level Comparison Reuse •SE shape independent digital hardware design approach•Principle: Comparisons along discrete shift levels (DSL)
are reused (orthogonal to the shift direction) while the SE is displaced line-wise over the image
OSLCR Architecture Principle
OSLCR: Orthogonal Shift Level Comparison Reuse •SE shape independent digital hardware design approach•Principle: Comparisons along discrete shift levels (DSL)
are reused (orthogonal to the shift direction) while the SE is displaced line-wise over the image
•Less comparisons per shift step -> enhance processing speed and hardware complexity
Architectural performance and hardware complexity analysis
Architecture Nand2 Gate Equ. relative max. frequency
Direct realization 14,976 1.0
This work (OSLCR) 6,576 6.17
Several SE shapes were realized in digital hardware (VHDL on register transfer level)
Implementation of erosion / dilation by OSLCR concept and direct realization
For e.g. diamond shaped 13 × 13 SE, 8 bit gray level input image
Compared to direct realization 44% of chip area
Approximately six times higher maximum clock frequency
Requirements concerning the transfer business
• We are looking for: Business Partners, Investors, Buyers of Licenses.
• We want to achieve: Research and Development Cooperation, Investment, Commercialization.
Please contact us for further information: [email protected], [email protected].
The End
Thank you for your attention…
Please contact us for further information: [email protected], [email protected].