Post on 09-Sep-2018
© 2014 HORIBA, Ltd. All rights reserved.
Introduction to Image Analysis
Jeffrey Bodycomb, Ph.D.HORIBA Scientific
www.horiba.com/us/particle
© 2014 HORIBA, Ltd. All rights reserved.
Size: Particle Diameter (m)0.01 0.1 1 10 100 1000
Colloidal
Suspensions and Slurries
DLS – SZ-100
Electron Microscope
Powders
Fine Coarse
Optical Microscopy PSA300, Camsizer
Laser Diffraction – LA-950
Acoustic Spectroscopy
Electrozone Sensing
Disc-Centrifuge
Light Obscuration
0.001
Macromolecules
Nano-Metric
Met
hods
App
sA
pps
Size
sSi
zes
Sedimentation
Sieves
© 2014 HORIBA, Ltd. All rights reserved.
Replace sieves (really!) Verify/supplement laser diffraction results (orthogonal
technique). Need shape information, for example due to
importance of powder flow
These may have the same size (cross section), but behave very differently.
Why image analysis?
© 2014 HORIBA, Ltd. All rights reserved.
More information available through www.retsch.com
Replace Sieves
Tend to wear over time. It is difficult to tell when sieve results are “drifting” due to wear
Results depend on nature of shaking and loading leading to operator to operator variations in results.
Small number of size classes
Why image analysis?
© 2014 HORIBA, Ltd. All rights reserved.
Need shape information for evaluating packing and flow.
Why image analysis?
© 2014 HORIBA, Ltd. All rights reserved.
Effect of Shape on Flow
Yes, I assumed density doesn’t matter. Roundness is a measure based on particle
perimeter.
c
© 2014 HORIBA, Ltd. All rights reserved.
Shape is important for roofing granules that block sunlight from reaching next layer.
Why image analysis?
© 2014 HORIBA, Ltd. All rights reserved.
Good Product
Compact particles that fully block UV
Flaky, angular, leaves gaps for UV to pass
BAD Product
Why image analysis?
© 2014 HORIBA, Ltd. All rights reserved.
Pictures: contaminants, identification, degree of agglomeration
Screen excipients, full morphology Root cause of error (tablet batches), combined
w/other techniques Replace manual microscopy
Why image analysis
© 2014 HORIBA, Ltd. All rights reserved.
Image Acquisitionand enhancement
Object/Phasedetection
Measurements
Major Steps in Image Analysis
© 2014 HORIBA, Ltd. All rights reserved.
Take a picture, analyze for size
Measurement Results
Image Analysis
© 2014 HORIBA, Ltd. All rights reserved.
Static Image Analysis
Particles are dispersed (isolated) on a surface
Picture are taken from stationary particles
Camera or surface with particles is shifted, multiple images are taken from different positions, images are processed and evaluated
High resolution images is possible
Number of images/particles is limited (because of time limitations)
Preferred orientation of the particles on the surface (largest 2D)
HORIBA PSA300
© 2014 HORIBA, Ltd. All rights reserved.
Dynamic Image Analysis
Particles flow through the measurement volume of the instrument and the field of view of the camera
Particles are captured during movement, no other moving parts necessary
Capturing of many particle images in a short time interval
Limitations because of image rate of the camera(s)
Image quality is (a bit) worse
Particles are projected in random orientation (3D) CAMSIZER
© 2014 HORIBA, Ltd. All rights reserved.
CCD - Basic CCD - Zoom
Measuring Principle
Detection of particlesOne pixel is element of a projection when at least half of the pixel is covered.
CAMSIZER XT 14µm 1µm
CAMSIZER
Resolution
© 2014 HORIBA, Ltd. All rights reserved.
Maximum size
Large particles cannot been measured properly even they fit in the frame.
© 2014 HORIBA, Ltd. All rights reserved.
Maximum size
The probability of large particles touching the edge of the frame is higher than for smaller particles.=> Large particles cannot been measured sufficiently
Upper limit of measurement range
© 2014 HORIBA, Ltd. All rights reserved.
Two-Camera-System
Basic-Camera Zoom-Camera
18© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
Acquiring Images
We want a good microscope and nice sharp images.
Pay attention to lighting and focus.
No Yes
© 2014 HORIBA, Ltd. All rights reserved.
CLEMEX
Stack images for sharper final image
Out of focus
Multilayer Grab for Sharpness
In focus
© 2014 HORIBA, Ltd. All rights reserved.
Dispersing a Sample
Want to spread particles out so that they don’t touch.
No Yes
© 2014 HORIBA, Ltd. All rights reserved.
Want to spread particles out so that they don’t touch.
Use % of field of view that is covered in order to control feed rate. Try 1% at first.
Control feed rate.
Feeding Too fast Good
© 2014 HORIBA, Ltd. All rights reserved.
Image Acquisitionand enhancement
Object/Phasedetection
Measurements
Major Steps in Image Analysis
© 2014 HORIBA, Ltd. All rights reserved.
Image Binarization
Turn into binary image (i.e., decide what is a particle and what isn’t).
© 2014 HORIBA, Ltd. All rights reserved.
Image Acquisitionand enhancement
Object/Phasedetection
Measurements
Major Steps in Image Analysis
© 2014 HORIBA, Ltd. All rights reserved.
xcmin
xc min
“width”
A
A‘ = A x are
a
“diameter overprojection surface”
xarea
“length”
xFe max
xFemax
Width is bestsuited for
comparison withsieves !
Shape parameterscan be calculated!
Many Size Measures
© 2014 HORIBA, Ltd. All rights reserved.
Comparison of Size Definitions
differentsize definitions
differentresults
xc min xArea xFemax
Measurement Results
x [mm]0.2 0.4 0.6 10
10
20
30
40
50
60
70
80
Q3 [%] Sample A_Basic_0.2%_xc_min_001.rdfSample A_Basic_0.2%_x_area_001.rdfSample A_Basic_0.2%_xFemax_001.rdf
2 x[mm]
© 2014 HORIBA, Ltd. All rights reserved.
comparison
CAMSIZER-measurement xarea (red)and sieving * (blue)
Digital Image Processing
Area Measurement Sieving
A
A‘ = A
x are
a
xarea“diameter
via projection surface”
29© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
x [mm]0.1 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Q3
Tinovetin-B-CA584A_BZ_xc_min_002.rdfSyngenta-1mm-2min-Sieb.ref
Digital Image Processing
Measuring of Width Sieving
Competing Measuring Methods
--- width measurement
-*- Sieving
comparison
CAMSIZER-measurement xc min (red)and sieving * (black)
xcmin
xc min
“width”
30© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
fitted result
CAMSIZER-measurement xarea (red)to sieving * (blue)
Fitting of CAMSIZER results to Sieving
31© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
Feret diam. 1
Longest diam.
Shortest diam.to longest┴
Aspect ratio
= shortest diamlongest diam
= to longest diamlongest diam
= shortest Feret diamlongest Feret diam
= three different numbers!
┴
Shape: Aspect Ratio
© 2014 HORIBA, Ltd. All rights reserved.
SPHT0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.900
10
20
30
40
50
60
70
80
90
Q3 [%]
#181-Conveyor_xc_min_002.rdf#181-Ballmilled_xc_min_001.rdf#231-Conveyor_xc_min_001.rdf
More round and compact
More irregular
PA2
4
A
P
Shape analysis
Back to those roofing granules
© 2014 HORIBA, Ltd. All rights reserved.
Data Evaluation
BinarizeFind Edges
Analyze Each Particle
Output Distribution
Raw Image
© 2014 HORIBA, Ltd. All rights reserved.
20,000 particles 200 particles
second populationmissed
“holes” in distribution
But d10, d50 &d90 may appear similar
Effect of Number of Particles Counted
© 2014 HORIBA, Ltd. All rights reserved.
More particles for more accuracy.
0
1020
3040
5060
7080
90
100
200
300
500
1000
2000
5000
1000
020
000
5000
010
0000
1771
87
dv10 dv50 dv90
Assume 49.833 is “correct”dv50,0.95 x 49.833 = 47.34 is within 95%,47.463 achieved by 5000 counts!
5000
Use this to control precision of your data (and not spend extra time on precision you don’t need.
© 2014 HORIBA, Ltd. All rights reserved.
Divide Large Data Set into Smaller Sets
Error bars are one standard deviation from repeated measurements of the same number of particles from different parts of the sample.
The error bars get smaller as you evaluate more particles.
© 2014 HORIBA, Ltd. All rights reserved.
<USP> 776: Standard Deviation
Implies normal particle size distribution, greater than 30 particles, and known standard deviation.
2
22
2
ba
nsns
=Moment of chi squared distribution (see a statistics book)s=estimated standard deviation of distribution (width)n=number of particles measured
These limits are asymmetric around the standard deviation.
© 2014 HORIBA, Ltd. All rights reserved.
How Many Particles?
Some materials have a distribution such that SD/Mean ~ 1. To obtain reliable mean values, measure ~1500 particles.To obtain more details about the distribution, (10x?) more particles need to be measured.
© 2014 HORIBA, Ltd. All rights reserved.
Accuracy
Is that the “real size”?
Image analysis uses actual pictures to extract size.
Calibration with a reticle.
© 2014 HORIBA, Ltd. All rights reserved.
Resolution
Measurement Results
mixture of six sizes of grinding balls42© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
Robust measurement….the interaction between the instrument and the particle is optical, so there is no wear and change in calibration.High resolution size distribution resultsFast
Also, these are all reasons to use Dynamic Image Analysis instead of sieves.
Why dynamic image analysis
© 2014 HORIBA, Ltd. All rights reserved.
The HORIBA PSA300
Turnkey SystemMore time getting results and
less time engineering
Automated Faster
Less operator labor
Less operator bias
Powerful Software Features Image Enhancement
Particle separation
Separate Disperser OptionMore flexible sample preparation
© 2014 HORIBA, Ltd. All rights reserved.
The CAMSIZER
Measurement of very broad particle distributions
Direct particle definition by width (analogue to sieving)
by length
or projection surface
Two camera system for more accuracy/wider range
Easy operation
Fail-safe, robust
Ideal for particle shape analyses
Measurement of density, counting of particles
Measurement Results
45© Retsch Technology GmbH
© 2014 HORIBA, Ltd. All rights reserved.
Static or Dynamic Image Analysis?
Dynamic
Broad size distributions (since it is easier to obtain data from a lot of particles)
Samples that flow easily (since they must be dropped in front of camera)
Powders, pellets, granules
Static
Samples that are more difficult to disperse (there are more methods for dispersing the samples)
Samples that are more delicate
Pastes, sticky particles, suspensions
© 2014 HORIBA, Ltd. All rights reserved.
Conclusions
Image Analysis is good for
Replacing Sieves
Size
Shape
Supplementing other techniques
Watch out for
Sample preparation
Image quality
Measure enough particles