Automatic morphometry and classification of crystals of iron oxides
An automatic method for the classification of crystals of iron oxides, including the measurement of oxide crystal size and shape, by a digital microscopy procedure involving image acquisition, processing, analysis, and classification is described.
1. An automatic method for classifying crystals of iron oxides comprising:
acquiring images of an iron ore, wherein the acquired images comprise a bright field image (BF) and one or more polarized light image in reflection mode; and
processing and analyzing the acquired images,
wherein the processing and analyzing of the acquired images comprises a synthetic step followed by an analytic step.
2. The method of claim 1 , wherein the polarized light image is captured in at least one of the modes selected from a group consisting of a linear mode (LPOL) and a circularly polarized light mode (CPOL).
3. The method of claim 2 , wherein in LPOL, an entry polarizer and an exit analyzer are positioned close to an extinction condition.
4. The method of claim 2 , wherein the acquiring includes acquiring an image in LPOL at a predetermined angle or a pair of images at symmetrical angles from an extinction position.
5. The method of claim 2 , wherein in CPOL, a single image is captured with circular polarization.
6. The method of claim 2 , wherein the one or more polarized light image is an image captured in the LPOL, an image pair captured in the LPOL, or an image captured in CPOL.
7. The method of claim 1 , wherein acquiring the images is performed by an optical microscope operating in a reflected light mode, including a fixed entry polarizer and a rotating exit polarizer or a system for circular polarization and a digital camera.
8. The method of claim 7 , wherein the optical microscope is computer controlled and motorized and the imaging mode (BF, LPOL, CPOL) is selected by software.
9. The method of claim 2 , comprising segmenting the BF by a threshold selection method to identify regions of hematite and preserving the regions of hematite.
10. The method of claim 9 , comprising applying the preserved regions of hematite as a mask to the one or more polarized light image captured in LPOL or CPOL to form one or more masked image.
11. The method of claim 10 , wherein the masked image is divided into square regions (textels) and Haralick parameters are measured for each textel.
12. The method of claim 11 , wherein the Haralick parameters are measured in 4 directions.
13. The method of claim 12 , wherein the 4 directions are 0, 45, 90, and 135 degrees.
14. The method of claim 1 , wherein the method is a reproducible method.
15. The method of claim 1 , wherein the method is independent of an operator.
16. The method of claim 1 , wherein the iron ore comprises hematite as a constituent mineral.
17. An automatic method for classifying crystals of iron oxides comprising:
acquiring one or more image of an iron ore; and
processing and analyzing the acquired one or more image,
wherein the processing and analyzing of the acquired one or more image comprises a synthetic step followed by an analytic step, wherein the synthetic step analyzes one or more image field of an acquired image and identifies the one or more image field as a morphology of iron oxide selected from a group consisting of granular, lamellar, lobular, microcrystalline, and matte.
18. The method of claim 17 , wherein the one or more image field is identified as granular, lamellar, or lobular and is submitted to the analytic step.
19. The method of claim 17 , wherein the synthetic step comprises measuring image texture based on Haralick parameters.
20. The method of claim 18 , wherein the analytic step comprises the detection of individual crystal boundaries by a Canny algorithm.
21. The method of claim 18 , wherein the analytic step uses a watershed algorithm.
22. The method of claim 21 , wherein closed contours formed by the watershed algorithm undergoes ultimate erosion until they are transformed into single pixel seeds.
23. The method of claim 22 , wherein the seeds are used as an input to a modified region growing algorithm.
24. An automatic method for classifying crystals of iron oxides comprising:
acquiring one or more image, wherein the acquired one or more image comprises two polarized light images captured in a linear mode (LPOL) or an image captured in a circularly polarized light mode (CPOL); and
processing and analyzing the acquired one or more image, comprising a synthetic step followed by an analytic step, wherein the analytic step uses a watershed algorithm to form closed contours that undergo ultimate erosion until transformed into single pixel seeds, the single pixel seeds being used as an input to a modified region growing algorithm, wherein crystals of iron oxides are reconstructed based on positions of the single pixel seeds and on the acquired one or more image.
25. The method of claim 24 , wherein a spectral distance d g (x,y) is calculated between each pixel p(x,y) connected to a crystal g and a value of its seed p(x g ,y g ) in the CPOL image or in the two LPOL images.
26. The method of claim 25 , wherein when the acquired one or more image comprises two polarized light images captured in the LPOL, if distance, or a maximum between two distances is smaller than a given-threshold t, the pixel p(x,y) belongs to crystal g.
27. The method of claim 25 , wherein when the acquired one or more image comprises two polarized light images captured in the LPOL, the spectral distance is defined as:
d
g
(
x
,
y
)
=
Max
(
d
1
g
(
x
,
y
)
,
d
2
g
(
x
,
y
)
)
d
i
g
(
x
,
y
)
=
(
R
i
(
x
,
y
)
-
R
i
(
x
g
,
y
g
)
)
2
+
(
G
i
(
x
,
y
)
-
G
i
(
x
g
,
y
g
)
)
2
+
(
B
i
(
x
,
y
)
-
B
i
(
x
g
,
y
g
)
)
2
(
1
)
where i=1 or 2 and refers to the 2 LPOL images; R i (x,y), G i (x,y), B ( x,y) are RGB values of pixel p(x,y) in the LPOL image i; R i (x g ,y g ), G i (x g ,y g ), B i (x g ,y g ) are RGB values of pixel p(x g ,y g ) in image LPOL i.
28. The method of claim 25 , wherein when the acquired one or more image comprises a polarized light image captured in the CPOL, the spectral distance is based on a single image as:
d
g
(
x
,
y
)
=
(
R
(
x
,
y
)
-
R
(
x
g
,
y
g
)
)
2
+
(
G
(
x
,
y
)
-
G
(
x
g
,
y
g
)
)
2
+
(
B
(
x
,
y
)
-
B
(
x
g
,
y
g
)
)
2
where R(x,y), G(x,y), B(x,y) are RGB values of pixel p(x,y) in the CPOL image; and
R(x g ,y g ), G(x g ,y g ), B(x g ,y g ) are RGB values of pixel p(x g ,y g ) in the CPOL image.