Image and Video Processing with

MATLAB

Dr. Roland Michaely

Application Engineer

2014 The MathWorks, Inc.1

Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
2
MathWorks at a Glance
Earths topography
on a Miller
cylindrical
projection, created
with MATLAB and
Mapping Toolbox.

Headquarters: Europe: Worldwide training

Natick, Massachusetts U.S. France, Germany, Italy, and consulting
Netherlands, Spain, Sweden,
Other U.S. Locations: Distributors serving
Switzerland, United Kingdom
California; Michigan; more than 20 countries
Texas; Washington, D.C. Asia-Pacific:
Australia, China, India,
Japan, Korea
3
MathWorks Today

Revenues ~$750M in 2012

More than 60% from outside the U.S.
Privately held
2800 employees worldwide
More than 1 million users in 175+
countries

4
Key Industries

Aerospace and defense

Automotive
Biotech and pharmaceutical
Communications
Education
Electronics and semiconductors
Energy production
Financial services
Industrial automation
and machinery
Medical devices

5
Image and Video Processing with MATLAB
Key Take Aways

High-level language

Development environment

Technical computing platform

6
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
7
Applications: Image and Video Processing

Medical imaging
Surveillance
Robotics
Automotive safety
Consumer electronics
Geospatial computing
Machine vision
and more

8
Technical Computing Tasks

Access Explore and Discover Share

Files Reporting and
Data Analysis Documentation
and Modeling

Algorithm
Software Outputs for Design
Development

Code and Applications

Application
Development
Hardware Deployment

Automate
10
Lets perform some image processing?

Import and Explore

Image Enhancement - Noise Removal
Image Segmentation Isolate Objects
Image Analysis / Statistics - Label & Count Objects

11
Demo: Candy Counter

12
Algorithm Summary

Import and Explore Images

Image Enhancement
Morphological Operators

Image Segmentation
Thresholding
Color Segmentation

Image Analysis
Object properties and statistics

13
Image Segmentation Other Methods
Active Contours and Watershed

14
Image Enhancement
Deblurring and Noise Removal

15
Image Registration
2-D & 3-D Image Alignment
Align multimodal images

16
Image Processing with Live Video

Validate algorithms with live data

Integrate with low cost webcams
Webcam support in base MATLAB

17
Image Processing Apps:
Segment Tool, MorphTool, Image Registration App

File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/ 19

Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
20
Features are Critical to Computer Vision

Edge Corner Template SURF MSER

21
Typical Parts of a Computer Vision Algorithm

1. Image/video acquisition Image Acquisition Toolbox

2. Image/video pre-processing Image Processing Toolbox
3. Feature detection
4. Feature extraction
5. Feature matching Computer Vision
System Toolbox
6. Using features
Stabilization, mosaicking
Stereo image rectification
7. Feature classification Statistics Toolbox

22
Face Detection, People Detection, and
Background Subtraction

23
Demo: Face tracking with a webcam

+ =

24
Stereo Vision (14a)

Stereo calibration
Semi-global disparity matching
Better results than block matching
3D scene reconstruction from
disparity
New demos available
Estimate real distance to objects in
video
3D scene reconstruction
Pedestrian detection (14b)

29
OCR Optical Character Recognition (14a)

Highly requested in Frontlines

Support for English, Japanese
Users can download additional
language support
Shipping demo with text detection
and OCR workflow

32
 Camera Calibration App: Demo

Simplified workflow for

estimating camera
intrinsic and extrinsic
parameters

Removes the effects of

lens distortion from an
image

Automatically detects
checkerboard patterns

35
Computer Vision Apps on File Exchange
CascadeTrainGUI App, CV ImageRegistration App

File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/ 36

Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
38
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
39
Common Image Processing Challenges

Reading and writing to various file formats

Gathering and exploring images to gain insight
Create, test and debug algorithms

Refine algorithms, make them robust and re-usable

Processing large images with limited memory
Executing algorithms faster

Sharing results and generating reports

Creating applications
Deploy to other environment
40
Range of Programming Techniques

value
variable
structure
Data
(properties)

Explore Automate Re-use class

(methods)
Algorithm
function
script
command line

41
How are System objects different from
MATLAB functions?

System Objects provide the following unique functionality:

1. Implicit state management, indexing and buffering

2. Support for fixed-point arithmetic+

3. Support for C code generation*#

4. Available in MATLAB and Simulink

+ with Fixed-Point Toolbox or Simulink Fixed Point

* with MATLAB Coder or Simulink Coder
# no code generation with Phased Array System Toolbox as of R2012a
43
Video Processing in MATLAB Before
Load the entire video file and process it all at once
Memory

Batch
Source Video
Algorithm

And with System Objects

Load a frame and process it before moving on to the next frame
MATLAB Memory

Stream Stream
Source Processing

44
Video Processing in MATLAB Before
Video processing is difficult in MATLAB

myVid = mmreader(myvideofile.avi);
numFrames = myVid.NumberOfFrames;
numIter = 10;
opticalFlowIn = zeros([size(currentFrame) 5]);
opticalFlowOutput = zeros([size(currentFrame) numFrames]);

i = 1; Explicit state management

while i <= numFrames
opticalFlowIn(:,:,2:end) = opticalFlowIn(:,:,1:end-1);
opticalFlowIn(:,:,1) = read(myVid,i);

flow = opticalFlow(opticalFlowIn(:,:,1),opticalFlowIn(:,:,5),
horn-schunck,numIter,magitude-squared);

opticalFlowOutput(:,:,i) = flow;
Explicit indexing
i = i+1;
end
Needs a buffer
implay(opticalFlowOutput,30)

45
Video Processing in MATLAB After
Video processing is difficult in MATLAB

but system objects make it easier.

reader = video.MultimediaFileReader
reader.Filename = myvideofile.avi; Object
viewer = video.DeployableVideoPlayer initialization
optical = video.OpticalFlow
optical.Method = horn-schunck;
optical.OutputValue = Magitude-squared;
optical.ReferenceFrameDelay = 3;
optical.MaximumIterationCount = 10;

Code in the loop more simple

while ~isDone(reader)
currentFrame = step(reader); Implicit state management,
OF = step(optical, currentFrame); indexing and buffering
step(viewer, OF);
end Video player works in the loop

46
Summary System Objects

New stream processing capabilities in MATLAB

Common user interface across all algorithms

Bridge MATLAB and Simulink

Code generation in MATLAB & Simulink

48
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
49
Images & Processing Demands on the Rise

Moving towards 4k / 8k UHD capture and display.

Efficiency and performance are becoming crucial.

50
Solving Larger Problems

Two primary types of larger problems:

Problems that take too long for
one computer to solve
Problems with data that is too big
for one computer

Problems can be addressed by leveraging

additional hardware
Multi-core or multi-processor computer
Graphics Processing Unit (GPU)
Clusters and clouds

51
Using More Hardware

Built-in multithreading
Automatically enabled in MATLAB since R2008a
Multiple threads in a single MATLAB computation engine
www.mathworks.com/discovery/multicore-matlab.html

Parallel computing using explicit techniques

Perform MATLAB computations on GPUs
Multiple computation engines controlled by a single session
High-level constructs to let you parallelize MATLAB applications

52
GPU vs. CPU Comparison
Features CPU GPU
# of Threads 10s 100s 100s 1000s

Thread Scheduler Host Operating System Dedicated Low Latency

(i.e. Windows / Linux) Scheduler
Memory System Memory / Cached High Speed / Low Latency
(Off Chip) (On Chip)
Upgradability Hard New motherboard Easy New PCIe card
and processor
Large Data Capability Medium High

Scalability Clusters and Clouds Clusters and Clouds

vs.
54
GPU Acceleration for 43 functions

Use NVIDIA GPUs to accelerate 43 of the most

popular Image Processing Toolbox functions
Accelerate complex algorithms with large data
Integrate CUDA code for prototyping and testing

bwmorph imhist
bwlookup imnoise
corr2 imopen
edge imresize
histeq imrotate
imadjust imshow
imbothat imtophat
imclose imwarp
imdilate mean2
imerode medfilt2
imfilter padarray
imgradient rgb2gray

55
Criteria for Good Problems to Run on a GPU

Massively parallel:
Calculations can be broken into hundreds
or thousands of independent units of work
Problem size takes advantage of many GPU cores

Computationally intensive:
Computation time significantly exceeds CPU/GPU data transfer time

Algorithm consists of supported functions:

Growing list of Toolboxes with built-in support
www.mathworks.com/products/parallel-computing/builtin-parallel-support.html
Subset of core MATLAB for gpuArray, arrayfun, bsxfun
www.mathworks.com/help/distcomp/using-gpuarray.html#bsloua3-1
www.mathworks.com/help/distcomp/execute-matlab-code-elementwise-on-a-
gpu.html#bsnx7h8-1

56
Example: Solving 2D Wave Equation
GPU Computing

Solve 2nd order wave

equation using spectral
methods:
2 2 2
2
= 2+ 2

Run both on CPU and

GPU
Using gpuArray and
overloaded functions

58
Benchmark: Solving 2D Wave Equation
GPU Computing

CPU GPU
Grid Size Speedup
(s) (s)

64 x 64 0.05 0.11 0.4

128 x 128 0.14 0.11 1.3

256 x 256 0.83 0.12 7.2

512 x 512 4.40 0.24 18.

1024 x 1024 18.8 0.82 23.

2048 x 2048 75.0 3.67 20.

CPU: Intel Xeon W3550 (3.07GHz)

GPU: NVIDIA Tesla K20c
59
Working with Large Images

Block processing with blockproc

Automatically divides an
image into blocks for
processing
Reduces memory usage
Processes arbitrarily large
images
Reduced resolution data set
rsetwrite
Avoids memory demands in
visualizing large images

60
Demo: filtering a large Image
Using blockproc and Built-in Parallel Support

61
Independent Tasks or Iterations
Parallel for-loops

Ideal problem for parallel computing

No dependencies or communications between tasks
Examples: parameter sweeps, Monte Carlo simulations

Tim Tim
e e
blogs.mathworks.com/loren/2009/10/02/using-parfor-loops-getting-up-and-
running/
62
Demos: GPU Computing, Blockproc, Parallel

64
Demo: Brain Scan and GPU Computing
Summary

Found bottleneck with profiler

Run on gpu using

gpuArray and gather

Compared speed-up cpu vs gpu

using timeit and gputimeit

gpuBenchmark

65
 Noisy Image
Demo: Filtering a large Image
Summary

Enabled built-in support for

Parallel Computing Toolbox
in Image Processing Toolbox

Used pool of MATLAB workers Filtered Image

Run median filtering in parallel

using blockproc

http://hirise.lpl.arizona.edu/
From - NASA/JPL/University of Arizona
67
Demo: Candy Counter parallel for loop and batch
Summary

Ran loops on a pool of

MATLAB resources

Converted for-loop
into parfor-loop

Run function on worker using

batch

68
Take Advantage of Cluster Hardware

Offload computation:
Computer Cluster
Free up desktop
Access better computers Cluster

Scale speed-up:
Use more cores
Go from hours to minutes
MATLAB Scheduler
Desktop (Client)
Scale memory:
Utilize distributed arrays
Solve larger problems without re-coding algorithms

69
 Heart Transplant Studies
3-4 weeks reduced to 5 days

Flight Test Data Analysis

16x Faster

Mobile Communications Technology

Simulation time reduced from weeks to
hours, 5x more scenarios

Hedge Fund Portfolio Management

Simulation time reduced from 6 hours to 1.2
hours

70
Offload Computations with batch

Work Worker
MATLAB
Desktop (Client) Worker
Worker
Result

Worker

batch()

71
Offload and Scale Computations with batch

Work Worker
MATLAB
Desktop (Client) Worker
Worker
Result

Worker

batch(,'Pool',)

72
Distributing Large Data

11 26 41

12 27 42

13 28 43

14 29 44
Worker 15 30 45
MATLAB 16 31 46
Desktop (Client) 17 32 47
Worker
Worker 17 33 48

19 34 49

20 35 50
Worker
21 36 51

22 37 52

Remotely Manipulate Array

from Client Distributed Array
Lives on the Workers
73
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
74
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
75
Common Image Processing Challenges

Reading and writing to various file formats

Gathering and exploring images to gain insight
Create, test and debug algorithms

Refine algorithms, make them robust and re-usable

Processing large images with limited memory
Executing algorithms faster

Sharing results and generating reports

Creating applications
Deploy to other environment
76
Deploying Algorithms and Applications

Access Explore and Discover Share

Files Reporting and
Data Analysis Documentation
and Modeling

Algorithm
Software Outputs for Design
Development

Code and Applications

Application
Development
Hardware Deployment

Automate
77
Next Steps - Sharing Results from MATLAB

Publish reports

Package as an App

Deploy applications

Deploying C Code

78
Publish Reports

Automatically generate reports

Publish MATLAB files
Detailed comments
Code examples
Customize reports using
MATLAB Report Generator

79
Packaging and Sharing MATLAB Apps

MATLAB apps
Interactive applications to perform
technical computing tasks
Displayed in apps gallery

Included in many MATLAB products

Package your own app

Create single file for distribution
and installation into gallery
Packaging tool:
Automatically includes all necessary files
Documents required products

80
 Deploying Applications - MATLAB Compiler

Share applications

Creates desktop or web

software components
MATLAB Compiler

Supports full MATLAB

MATLAB MATLAB MATLAB
language and most Builder EX Builder JA Builder NE

toolboxes

Requires MATLAB
Compiler Runtime
Free run-time library .exe .dll Excel Java Web COM .NET

Royalty-free deployment

81
 Deploying C Code - MATLAB Coder

Generate C code

High performance C code

with shared libraries
MATLAB Coder
Integrate with existing
software

Royalty-free deployment
.c .lib

Supports subset of the Intel x86 /x64

MATLAB language and
some toolboxes

Intel x86 / x64

82
MATLAB Compiler and MATLAB Coder

MATLAB Compiler MATLAB Coder

.exe .dll .c

.exe .lib .dll

83
 MATLAB Compiler or MATLAB Coder?

The Middle Ground

.c

MATLAB Coder
MATLAB Compiler
Readable
Builders, MPS Desktop
Portable

.dll .c
.cpp
Web .exe
Excel

.NET

Java

.exe

84
Code Generation for 41 IPT Functions

padarray imhmax
bwmorph imhmin
bwlookup imreconstruct
conndef imregionalmax
fspecial iptcheckconn
imcomplement label2rgb
imfill

Only for PC platforms Intel & AMD devices

85
Code Generation for ALL 86 CVST Functions

Recently added functions that use OpenCV

Support now includes
vision.ForegroundDetector
extractFeatures
matchFeatures
detectSURFFeatures
vision.CascadeObjectDetector

86
Why Engineers Translate MATLAB to C Today

.exe Prototype MATLAB algorithms on desktops as

standalone executables

.lib Integrate MATLAB algorithms with existing C

.dll environment using source code and static/dynamic
libraries

.c

+ Implement highly optimized C code and Libraries on

Intel processors
.lib

MEX Accelerate user-written MATLAB algorithms

87
Automatic Translation of MATLAB to C
.c
e
rat
i te
.exe

.lib
Algorithm Design and .dll
Code Generation in
verify
MATLAB a cce /
lerate
MEX

With MATLAB Coder, design engineers can:

Maintain one design in MATLAB

Design faster and get to C quickly
Test more systematically and frequently
Spend more time improving algorithms in MATLAB

88
Choosing the Right Deployment Solution
MATLAB Coder and MATLAB Compiler

.c

MATLAB Compiler MATLAB Coder

Executable or software Portable and readable

Output component/library C source code

Full language
Subset of language
MATLAB support Most toolboxes
Some toolboxes
Graphics

MATLAB Compiler
Runtime requirement Runtime
None

License model Royalty-free Royalty-free

90
Deployment: Demos
MATLAB Compiler, MATLAB Coder, and Apps

91
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
92
Target External Hardware
Code Generation Technologies
Embedded Coder
Automatically generate C and C++
Embedded optimized for embedded systems
Coder comparable to the efficiency of
handwritten code
Simulink Coder
Automatically generate C and C++ from
Simulink
Coder Simulink models and Stateflow charts
for Rapid Prototyping and Hardware-in-
the-Loop

MATLAB MATLAB Coder

Coder
Automatically generate C and C++ from
the suitable MATLAB subset

93
 Target External Hardware
Simulink Hardware Support

Arduino Lego NXT Raspberry Pi

BeagleBoard PandaBoard Gumstix Overo

Selected Simulink Supported Target Hardware:

Raspberry Pi Model B ($40)
Arduino Uno, Mega 2560, Nano, Ethernet Shield ($30-$70)
LEGO MINDSTORMS NXT ($280)
BeagleBoard-xM ($150)
http://www.mathworks.com/hardware-support/home.html PandaBoard ($180)
Gumstix Overo hardware ($180)

Available in Student Version!

94
Target External Hardware
Example Raspberry Pi

Raspberry Pi hardware
support package
Capturing live data from
sensors and imaging devices
(camera board)
Connects to MATLAB over
Ethernet (wired or wireless)

95
Raspberry Pi Hardware Setup

Connect Raspberry Pi
Plug in USB Ethernet Adapter to PC,
and Ethernet cable into adapter and
Raspberry Pi
Plug USB webcam into Raspberry Pi
Insert SD card (which may be a carrier
for SD micro) into Raspberry Pi
Attach USB power cable USB A to
PC, and USB micro B to power port on
Raspberry Pi
Wait a minute for Raspberry Pi to boot

Lets go test it! 96

Raspberry Pi Demos

Object Detection
Locate and mark the center of green object

Barcode scanner
Scan barcode and output numerical digits

97
Agenda
Welcome and Introductions
Image Processing with MATLAB
Computer Vision with MATLAB
Break
Programming Techniques
Speeding-up your Applications
Break
Deploy your Applications
Target External Devices
Summary
98
Image and Video Processing with MATLAB
Key Take Aways

High-level language
Native support for vector and matrix operations
Built-in math and visualization functions

Development environment
Interactive and easy to get started
Ideal for iterative exploration and design

Technical computing platform

Add-on products for a range of application areas
(e.g., signal processing and communications, image and video
processing, control systems, test and measurement)

99
Why use MATLAB for Image Processing?

Read and write many image file formats

Visualize and explore images interactively
Work with live video cameras
Use a large library of built-in functions
Quickly build custom image processing algorithms
Block-process large images to avoid memory issues
Process images faster with multiple cores and GPUs

100
Image Processing Toolbox

Perform image processing, analysis, visualization, and

algorithm development

Image analysis
Image enhancement
Geometric transformation
Image registration
Morphological operations
ROI-based processing
Image display and exploration

101
Why Use MATLAB for Computer Vision?

Increased productivity over C/C++ programming

Faster to prototype
More options with the camera
Comprehensive environment
Analysis, algorithm development, visualization, etc.
Broad library of algorithms
Computer vision
Image processing
Classification and clustering
Control system design, optimization, code generation
Documentation, examples, and technical support
Integrated workflow from design to implementation
102
Computer Vision System Toolbox

Design and simulate computer vision

and video processing systems

Feature detection and extraction

Registration and stereo vision
Object detection and tracking
Motion estimation
Video processing, file I/O, display,
and graphic overlays

103
Image Acquisition Toolbox

Acquire images & video from industry-standard hardware:

Camera Link
DCAM compatible FireWire (IIDC 1394)
GigE Vision
GenICam Interface

Manufacturers include:
Allied Vision Technologies
Basler
Baumer
FLIR
Hamamatsu
Point Grey
Teledyne DALSA
See Hardware Catalog for more details: www.mathworks.com/hardware
104
Support and Community

105
 Consulting Services
Accelerating return on investment

A global team of experts supporting every stage of tool and process integration

Process and Technology

Continuous Improvement
Automation
Process and Technology
Standardization

Full Application
Deployment
Process Assessment
Component
Deployment
Advisory Services

Jumpstart
Migration Planning

Research Advanced Engineering Product Engineering Teams Supplier Involvement

106
Training Services
Exploit the full potential of MathWorks products

Flexible delivery options:

Public training available worldwide
Onsite training with standard or
customized courses
Web-based training with live, interactive
instructor-led courses
Self-paced interactive online training

More than 30 course offerings:

Introductory and intermediate training on MATLAB, Simulink,
Stateflow, code generation, and Polyspace products
Specialized courses in control design, signal processing, parallel
computing, code generation, communications, financial analysis, and
other areas

107
 MATLAB Central

Community for MATLAB and Simulink users

70k daily visits

File Exchange
Access more than 10k free files
including functions, apps, examples,
and models

MATLAB Answers
Ask programming questions or search
18k+ community-answered Questions

Newsgroup
Participate in technical discussions
1,400 posts per day

Blogs
Read commentary from engineers who
design, build, and support MathWorks products

Based on average 2011 data 108

Connections Program

More than 400 add-on products and services that

complement and extend MathWorks products:
Third-party toolboxes and libraries for MATLAB and Simulink
Interfaces to third-party software and hardware products
Specialized training courses and consulting services
System integrators that incorporate MathWorks products

MATLAB and
Simulink interfaces, Embedded processors, real-
toolboxes and libraries time systems and
instrumentation
109
Book Program

More than 1400 books for educational and professional

use, in 28 languages
Control systems Mathematics
Digital signal processing Physics
Image & video processing Systems modeling
Biosciences & biomedical Chemistry
Communications systems Computational finance
Mechanical engineering Electronics

110
From a Leading Textbook Author ...

"I have used a number of commercial

image processing packages over the
years, and prefer the MathWorks
Image Processing Toolbox for several
reasons: the wide variety of functions it
provides, the users ability to write
additional functions with minimal effort,
the quality of the software, and the high
level of support."

Rafael C. Gonzalez
Professor Emeritus
University of Tennessee

111
Technical Support
Resources High customer satisfaction
Over 100 support engineers 95% of calls answered
All with MS degrees (EE, ME, CS) within three minutes
Local support in North America, 70% of issues resolved
Europe, and Asia within 24 hours
Comprehensive, product-specific Web 80% of customers surveyed
support resources rate satisfaction at 80100%

112
For More Information

Experiment with product by downloading a trial

Peruse videos, webinars, user stories, and demos
online
Contact us
Talk to a sales representative
to get answers to your questions
Discuss your projects
with MathWorks applications
engineers

113
Thank You for Attending Todays Seminar

Pete Tsinzo Customer Service

Account Manager info here
Add relevant Email: [email protected]
Email: [email protected] Phone: 508.647.7000 option 1
Phone: 310-819-4996

MathWorks Technical Support

www.mathworks.com Email: [email protected]
Phone: 508.647.7000 option 2

115
Dr. Roland Michaely
Application Engineer

2014 The MathWorks, Inc.

116