DSP-FPGA Based Image Processing System

DSP
-
FPGA Based Image
Processing System


Checkpoint Presentation

Jessica Baxter


Sam Clanton

Simon Fung
-
Kee
-
Fung

Almaaz Karachi


Doug Keen

Computer Integrated Surgery II

April.19.2001

Plan of Action

•
Project Description and Deliverables

•
Implementation Overview

•
Progress to Date

•
Problems Encountered

•
Dependencies

Project Overview

•
Objective
: To develop a robust image
processing system using adaptive image
segmentation, taking advantage of a DSP
and FPGA hardware implementation to
increase speed.

•
Deliverables
:

–
Minimum: Adaptive Image Segmentation Software

–
Expected: Software Implemented in Hardware,
Handling of Static Images

–
Maximum: Real
-
time Handling of Live Input

Plan of Action

•
Project Description and Deliverables

•
Implementation Overview

•
Progress to Date

•
Problems Encountered

•
Dependencies

Adaptive Image Segmentation
System: Software Side

1)
Input image

2)
Compute image statistics.

3)
Segment the image using initial
parameters.

4)
Compute the segmentation quality
measures

5)
WHILE not <
stop conditions
> DO

a)
Select individuals using the
reproduction operator

b)
Generate new population
using the crossover and
mutation operators

c)
Segment the image using
new parameters

d)
Compute the segmentation
quality measures

END

6)
Update the knowledge base using
the new knowledge structure

Figure: Bhanu, Lee

Hardware Assignment



•
The DSPs will serve as the main processor
and the FPGAs will provide support as co
-
processors.


Functional Break
-
Up:

FPGA:

•

Image Acquisition

•
Basic Image Processing (ex.
Brightness)

•
Image Analysis
–

choosing and
calculating statistical parameters

•
Segmentation
–

must also create
a vector graphic file for the
segmented data

•
Evaluation of Metrics of
Population Fitness

DSP:

•
Initiation of Genetic Algorithm

•
Optimization

•
Join
–

calls vector graphic file to
align segmented pieces

CRT:

•

Output (including values of
statistical evaluation parameters)


Back to the Plan

•
Project Description and Deliverables

•
Implementation Overview

•
Progress to Date

•
Problems Encountered

•
Dependencies

Key Dates

•
March 5
th

-

Topical research should be completed and we
should have a hashed out algorithm approved by Dr. Bey


•
March 12
th

–

Development Platforms determined, and
development framework in place.

•
March 19
th

–

CHANGE OF PROJECT OBJECTIVE
–

Combining
of DSP and FPGA

implementations

•
April 2
nd

–

Assign programming responsibilities to each
individual and hash out new algorithm

•
April 9
th

–

Troubleshoot the programming outline.

•
April 16
th

–

Outline of code, Interfaces between functions

•
April 23
rd

–

Integrate program components, test, and debug

•
April 25
th

–

Drive to hardware and test on static images

•
April 27
th

–

Refined

Image Analysis:

The Ins and Outs

•
In
-

From Preprocess

–
Image Struct

•
Height

•
Width

•
Raw Data

•
Output
-

To GA

–
Array

•
Statistics


Statistics

•
Mean

•
Variance

•
Centroid

•
Skewness

•
Energy

•
Entropy

•
Kurtosis


•
N(v) is
histogram


•
P(v) = N(v)/S


•
S = Image
Size

How?

Progress Status for Image Analysis

•
Analysis Problems

–
MATLAB

–
Test Data


•
Status

–
95% done

–
To do: More testing

•
Further Improvements

–
More Parameters

–
Even More Parameters


Evaluation

•
Measure overall quality of image
segmentation

•
Compare edginess of foreground with
edginess of background


Fitness
E
E
E
E
E
E
otherwise
background
foreground
f
b
foreground
background









Best Image Processing

•
Optimization problem

•
“Twiddling Knobs” Approach

•
Genetic Algorithm Approach

Figure: Bhanu, Lee

GA method for Image
Segmentation

Figure: Bhanu, Lee

Flow of Genetic Adaptation Cycle

•
Cycle:

–
Segmentation

–

Evaluation

–
Reproduction

–
Recombination

•
Cycle continues until
acceptable
segmentation results
are achieved

•
Long term pop. Is
then modified in
order to retain the
information “learned”
during the GA
process

Figure: Bhanu, Lee

Evolution of Segmentation
System

Figure: Bhanu, Lee

Background Extraction

•
Extract background from input image to
isolate areas that contain useful information

•
Use algorithm presented in:


Rodriguez, Arturo A., Mitchell, O. Robert. “
Robust

statistical method for background extraction in

image segmentation
”
Stochastic and Neural

Methods in Signal Processing, Image

Processing, and Computer Vision
. Vol. 1569,

1991

•
Output to evaluation module


Background Extraction

Background Extraction

Inputs

•
Image

•
Parameters necessary
for background
extraction algorithm

•
Parameter indicating
edge detection
algorithm and that
algorithm’s parameters

Output

•
Three
-
Layer Image
Array

–
1
st

Layer: Original Image

–
2
nd

Layer:
Background/Foreground
Image
(black=background,
white=foreground)

–
3
rd

Layer: Edge image

Background Extraction:
Progress/Difficulties


•
Many algorithmic details left out in research
paper

•
Mostly implemented

•
Debugging phase

Once Again…The Plan

•
Project Description and Deliverables

•
Implementation Overview

•
Progress to Date

•
Problems Encountered

•
Dependencies

Problems Encountered

•
Change in Project Objective: Integration of
both hardware types into one system rather
creating 2 hardware systems

•
Conversion of data types from C


Matlab

•
Testing of quality measure statistics during
image analysis

•
Determination of the Threshold for Stopping
the GA (i.e. Fitness Evaluation) is rather
subjective

•
Porting to Hardware (also, compatibility of
code with hardware capabilities)

And Finally…


•
Project Description and Deliverables

•
Implementation Overview

•
Progress to Date

•
Problems Encountered

•
Dependencies

Dependencies



Solved!


•
DSP and FPGA hardware obtained from TI
and Xilinx, respectively

•
Xilinx software obtained to drive code down
to hardware level



Still Waiting On…

•
Image Capture Device
-

Important for
reaching the maximum goal of real
-
time
visual processing

•
Assembly of hardware components


fin