Claims:I/WE CLAIM:
1. A method for identifying and estimating ferrite, martensite and bainite phases from nital etched images of dual phase steel, the method comprising:
taking nital images where sample regions are marked, wherein the natal images are raw steel images (100) obtained from a microscope;
applying an edge removal process on the natal images to obtain edge removed images (101);
extracting features (102) from a region of interest surrounding each marked pixel of the sample regions marked in the natal images, wherein the extracted features (102) include a histogram of contour lines, a histogram of entropy, and a histogram of mean-variance of each marked pixel;
concatenating the extracted features (102) to form a feature vector corresponding to each marked pixel;
training a random forest classifier (104) with the feature vectors and corresponding labels for phases, wherein the labels for phases are extracted from already stored ground truth images of the phases;
testing the trained random forest classifier (104) based on the training by taking nital test images where a test is to be performed without marked regions, wherein nital test images are test images (200) of raw steel image obtained from a microscope;
extracting features (202) from a region of interest surrounding each pixel of the test image (200), wherein the extract features (202) include a histogram of contour lines, a histogram of entropy, and a histogram of mean-variance of each pixel of the test image (200);
concatenating the extracted features (202) of each pixel of the test image to form a feature vector corresponding to each pixel of the test image (200);
feeding the feature vectors to the trained random forest (104); and
providing, using the trained random forest (104), a class label corresponding to the phase of each pixel as an output image (203).
2. The method as claimed in claim 1, wherein the edge removed image contains whitish ferrite region and non-white greyish region representing both bainite and martensite regions, and wherein after removing the edges, the method comprises subjecting non-white greyish region to two class classification by using random forest-based ensemble classification.
3. The method as claimed in claim 1, wherein ferrite, bainite, and martensite are identified based on different colour attributed to the microstructure.
4. The method as claimed in claim 1, wherein the edge removal process comprises:
binarizing the collected raw steel image (100) to get a binary image, wherein the collected raw steel image (100) comprises regions with edges, the regions including ferrite regions, bainite regions, and martensite regions;
performing dilation on the binary image to get a dilated image by removing boundary pixels of the bainite regions and the martensite regions;
subjecting the dilated image to erosion to obtain an eroded image; and
masking the eroded image by manipulating black and white pixels to obtain the edge removed image (101).
5. The method as claimed in claim 4, wherein masking of the eroded image comprises keeping unaltered black pixels of the eroded image and masking remaining pixels of the eroded image so as to obtain the edge removed image (101).
6. The method as claimed in claim 4, wherein masking of the eroded images is performed using standard image processing techniques.
7. The method as claimed in claim 6, wherein the standard image processing techniques include image morphological operations.
8. A method for testing a random forest classifier (104), comprising:
taking nital test images where a test is to be performed without marked regions, wherein nital test images are test images (200) of raw steel image obtained from a microscope;
removing edges of the test image (200) by using an edge removal process;
extracting features (202) from a region of interest surrounding each pixel of the test image (200), wherein the extract features (202) include a histogram of contour lines, a histogram of entropy, and a histogram of mean-variance of each pixel of the test image (200);
feeding the feature vectors to the trained random forest (104);
providing, using the trained random forest (104), a class label corresponding to the phase of each pixel as an output image (203); and
performing analysis (204) of the output image (203).
9. The method as claimed in claim 8, further comprising testing the edge removed images using the k-fold cross-validation, by:
performing testing the images in one fold and training with the images in the other (k-1) folds;
performing five-fold cross-validation on a non-ferrite region, corresponding to bainite and martensite pixels, of twenty images;
performing training of a random forest classifier (104) using sixteen images out of the twenty images; and
testing sequentially only four images out of the twenty images.
10. The method as claimed in claim 8, wherein the output image (203) is generated using mean and variance feature.
11. The method as claimed in claim 8, wherein the output image (203) is generated using multilevel mean and standard deviation. , Description:TECHNICAL FIELD
[0001] The present disclosure relates to a method for identification and estimation of ferrite, bainite and martensite phases from nital etched images of dual phase steel.
BACKGROUND
[0002] Background description includes information that may be useful in understanding the present subject matter. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed subject matter, or that any publication.
[0003] Dual-phase (DP) steel has a microstructure consisting of martensite and/or bainite phases embedded in the ferrite matrix. Though the second phases (martensite/bainite) have a distinct appearance when compared to the ferrite matrix, little contrast exists between martensite and bainite phases. However, as a consequence of different mechanisms of formation, a morphological difference between these two second phases do exist.
[0004] Commercialized image analyzer packages available with optical microscopes are not able to distinguish between martensite and bainite. The images from the optical microscopes provide a great problem for calculating phase fractions of these phases from a micrograph leaving no option but to rely on linear intercept method which is a very crude method.
[0005] The appearance of different phases of steel microstructure may vary in different samples. It is better to rely on more fundamental properties of the phases for feature extraction. Based on the formation process, features of different phases of the nital etched images are evaluated. The dual-phase steel formation process starts from low or medium carbon steel (ferrite and austenite). The carbon steel has undergone a process through continuous cooling, which produces the results of an atomic rearrangement. The atomic rearrangement is termed as phase transformation.
[0006] Martensite and bainite phases are formed due to phase transformation. But the procedure of formation of martensite and bainite are different. If diffusion of carbon occurs through iron, bainite is formed. If carbon is not diffused during phase transformation and the iron-carbon solid solution undergoes a sudden diffusion-less shear process, martensite is formed. These differences in the physics of formation result in differences in the intensity patterns of images of different phases.
[0007] Ferrite contains a negligible amount of carbon. So, ferrite has a whitish and homogeneous appearance. Once a sufficient amount of ferrite is produced, the steel is cooled rapidly. Due to the rapid cooling, carbon does not diffuse and a sudden diffusion-less sheer process takes place. This sheer process changes the atomic arrangement of the material and martensite is formed. Due to the presence of carbon, martensite regions are generally darker and less homogeneous than ferrite. During martensite formation, some amount of austenite (which is not sufficiently enriched with carbon) may be transformed to bainite as well. Also, the formation of bainite is diffusionless. However, after bainite formation, carbon diffuses out and fine carbides are formed. Further, due to these carbides, bainite is the most non-homogeneous and darkest among the three phases.
[0008] Moreover, the microscopic images of dual phase steel contains lots of edges in between the ferrite regions. Bainite and Martensite regions are blackish, Ferrite regions are whitish and edges are dark (blackish) in nature. So, the pixels in the edges have similarity with the Bainite and Martensite pixels. When the images are classified into the image pixels, the pixels in the edges and some of its neighboring pixels are classified into either Bainite pixels or Martensite pixels. Because of this, the phase fraction of Ferrite was decreasing and phase fractions o] f Bainite and Martensite were increasing.
[0009] However, the conventional method by using the optical microscope has many problems for calculating phase fractions of these phases from a micrograph and linear intercept method which is a very crude method.
[0010] Hence, there is a need a method in such a manner, that, the above-mentioned drawbacks of the conventional method are addressed.
OBJECTS OF THE DISCLOSURE
[0011] Some of the objects of the present disclosure, which at least one embodiment herein satisfy, are listed hereinbelow.
[0012] A general object of the present disclosure is to implement a method for calculating the phase fraction of martensite and bainite from a given optical and/or scanning electron micrograph by which it can easily distinguish the two phases by using nital etched microscopic images of dual phase (DP) steel.
[0013] It is an object of the present disclosure is to overcome the aforementioned and other drawbacks in prior method/product/apparatus.
[0014] It is another object of the present disclosure is to develop a method to identify ferrite, martensite, and bainite as separate phases in optical and/or scanning electron micrograph by image analysis.
[0015] It is another object of the present disclosure is to develop a method to estimate phase fraction of ferrite, martensite and bainite from optical micrograph for a microstructure comprising of martensite and/or bainite phases as second phases embedded in a ferrite matrix.
[0016] It is yet another object of the present disclosure is to develop a method user-friendly.
[0017] These and other objects and advantages of the present disclosure will be apparent to those skilled in the art after a consideration of the following detailed description taken in conjunction with the accompanying drawings in which a preferred form of the present disclosure is illustrated.
SUMMARY
[0018] This summary is provided to introduce concepts related to a method for identification and estimation of ferrite, bainite and martensite phases from nital etched images of dual phase steel. The concepts are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.
[0019] The present disclosure relates to a method for identifying and estimating ferrite, martensite and bainite phases from nital etched images of dual phase steel. The method includes taking nital images where sample regions are marked, wherein the natal images are raw steel images obtained from a microscope; applying an edge removal process on the natal images to obtain edge removed images; extracting features from a region of interest surrounding each marked pixel of the sample regions marked in the natal images, wherein the extracted features include histogram of contour lines, histogram of entropy, and histogram of mean-variance of each marked pixel; concatenating the extracted features to form a feature vector corresponding to each marked pixel; training a random forest classifier with the feature vectors and corresponding labels for phases, wherein the labels for phases are extracted from already stored ground truth images of the phases; testing the trained random forest classifier based on the training by taking nital test images where a test is to be performed without marked regions, wherein nital test images are test images of raw steel image obtained from a microscope; extracting features from a region of interest surrounding each pixel of the test image, wherein the extract features include histogram of contour lines, histogram of entropy, and histogram of mean-variance of each pixel of the test image; concatenating the extracted features of each pixel of the test image to form a feature vector corresponding to each pixel of the test image; feeding the feature vectors to the trained random forest; and providing, using the trained random forest, class label corresponding to the phase of each pixel as an output image.
[0020] In an aspect, the edge removed image contains whitish ferrite region and non-white greyish region representing both bainite and martensite regions, and wherein after removing the edges, the method comprises subjecting non-white greyish region to two class classification by using random forest-based ensemble classification.
[0021] In an aspect, ferrite, bainite, and martensite are identified based on different colour attributed to the microstructure.
[0022] In an aspect, the edge removal process comprises binarizing the collected raw steel image to get binary image, wherein the collected raw steel image comprises regions with edges, the regions including ferrite regions, bainite regions, and martensite regions; performing dilation on the binary image to get dilated image by removing boundary pixels of the bainite regions and the martensite regions; subjecting the dilated image to erosion to obtain an eroded image; and masking the eroded image by manipulating black and white pixels to obtain the edge removed image.
[0023] In an aspect, masking of the eroded image comprises keeping unaltered black pixels of the eroded image and masking remaining pixels of the eroded image so as to obtain the edge removed image.
[0024] In an aspect, masking of the eroded images is performed using standard image processing techniques.
[0025] In an aspect, the standard image processing techniques include image morphological operations.
[0026] The present disclosure further relates to a method for testing a random forest classifier. The method includes taking nital test images where a test is to be performed without marked regions, wherein nital test images are test images of raw steel image obtained from a microscope; removing edges of the test image by using edge removal process; extracting features from a region of interest surrounding each pixel of the test image, wherein the extract features include histogram of contour lines, histogram of entropy, and histogram of mean-variance of each pixel of the test image; feeding the feature vectors to the trained random forest; providing, using the trained random forest, class label corresponding to the phase of each pixel as an output image; and performing analysis of the output image.
[0027] In an aspect, the method further includes testing the edge removed images using the k-fold cross-validation, by performing testing the images in one fold and training with the images in the other (k-1) folds; performing five-fold cross-validation on a non-ferrite region, corresponding to bainite and martensite pixels, of twenty images; performing training of a random forest classifier using sixteen images out of the twenty images; and testing sequentially only four images out of the twenty images.
[0028] In an aspect, the output image is generated using the mean and variance feature.
[0029] In an aspect, the output image is generated using multilevel mean and standard deviation.
[0030] It is to be understood that the aspects and embodiments of the disclosure described above may be used in any combination with each other. Several of the aspects and embodiments may be combined to form a further embodiment of the disclosure.
[0031] These and other objects and advantages of the present invention will be apparent to those skilled in the art after a consideration of the following detailed description taken in conjunction with the accompanying drawings in which a preferred form of the present invention is illustrated.
BRIEF DESCRIPTION OF THE DRAWINGS
[0032] It is to be noted, however, that the appended drawings illustrate only typical embodiments of the present subject matter and are therefore not to be considered for limiting of its scope, for the invention may admit to other equally effective embodiments. The detailed description is described with reference to the accompanying figures. In the figures, a reference number identifies the figure in the reference number first appears. The same numbers are used throughout the figures to reference like features and components. Some embodiments of system or methods or structure in accordance with embodiments of the present subject matter are now described, by way of example, and with reference to the accompanying figures, in which:
[0033] FIG. 1A illustrates the block diagram for training a random forest classifier according to an embodiment of the present disclosure;
[0034] FIG. 1B illustrates the block diagram for testing a random forest classifier according to an embodiment of the present disclosure;
[0035] FIG. 2 illustrates the pictorial view of picking of neighbourhood of a pixel with removed edges image according to an embodiment of the present disclosure;
[0036] FIG. 3 illustrates the pictorial view of the mean-variance histogram curve; in accordance with an embodiment of the present disclosure
[0037] FIG. 4 illustrates a pictorial view of the original image and with the cropped region of the original image with edges; in accordance with an embodiment of the present disclosure;
[0038] FIG. 5 illustrates a flow chart of edge removal process steps in which the edge from the sample region of ferrite, martensite, and bainite are removed during training and/or testing of a random forest classifier according to an embodiment of the present disclosure;
[0039] FIG. 6 illustrates a pictorial view of the original image and edge removed image after edge removal process according to an embodiment of the present disclosure;
[0040] FIG. 7 illustrates a mean-variance curve of martensite and bainite pixels represented as red points and green points wherein, red points represent martensite pixels, and green points represent bainite pixels;
[0041] FIG. 8A illustrates a multilevel mean-standard deviation curve of martensite and bainite pixels represented as red points and green points wherein, red points represent martensite, and green points represent bainite;
[0042] FIG. 8B illustrates a multilevel mean-standard deviation curve for bainite class only represented in green points only;
[0043] FIG. 8C illustrates a multilevel mean-standard deviation curve for martensite class only represented in red points only;
[0044] FIG. 9A illustrates the output image generated by using mean and variance feature;
[0045] FIG. 9B illustrates an output image generated by using multilevel mean and standard deviation; and
[0046] FIG. 10 illustrates the description of Multilevel Mean and Standard Deviation calculation.
[0047] The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION
[0048] The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.
[0049] It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
[0050] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises”, “comprising”, “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
[0051] It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
[0052] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
[0053] In the present subject matter, it is directed to a method for identifying and estimating ferrite, martensite and bainite phases from Nital etched images of dual phase steel. The method comprises two steps as training and testing of a sample region. First, a set of Nital images are used for training an ensemble classifier, random forest and different portion of these nital images are manually marked for bainite, martensite and ferrite regions. Second, the trained random forest is used to test a nital-etched microscopic image of steel by extracting various essential features of sample region, once after edge removal process is done.
[0054] In the embodiment, once the images of dual phase steel are provided from the microscope, the image contains lots of edges, in between the ferrite regions, Bainite and Martensite regions, which are dark (blackish) in nature. Ferrite regions are whitish and Bainite and Martensite regions are blackish. The pixels in the edges have similarity with Bainite and Martensite pixels. While classifying the image pixels, the pixels in the edges and some of its neighbouring pixels are classified into either Bainite pixels or Martensite pixels. Finally, the phase fraction of Ferrite is decreased and the phase fraction of Bainite and Martensite are increased. To overcome this problem, there is a need of edge removal process, wherein the edges from the original images are removed by using an image processing technique. Then, the testing on the new edge removed images are performed. As a result, the phase fraction of Ferrite is increased and the phase fraction of Bainite and Martensite are decreased.
[0055] It should be noted that the description and figures merely illustrate the principles of the present subject matter. It should be appreciated by those skilled in the art that conception and specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present subject matter. It should also be appreciated by those skilled in the art that by devising various arrangements that, although not explicitly described or shown herein, embody the principles of the present subject matter. Furthermore, all examples recited herein are principally intended expressly to be for pedagogical purposes to aid the reader in understanding the principles of the present subject matter and the concepts contributed by the inventor(s) to furthering the art and are to be construed as being without limitation to such specifically recited examples and conditions. The novel features which are believed to be characteristic of the present subject matter, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures.
[0056] These and other advantages of the present subject matter would be described in greater detail with reference to the following figures. It should be noted that the description merely illustrates the principles of the present subject matter. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described herein, embody the principles of the present subject matter and are included within its scope.
[0057] Referring to FIG. 1A and FIG. 1B illustrates a method for identifying and estimating ferrite, martensite and bainite phases from nital etched images of dual phase steel. The raw steel images (100) are taken from the microscope. The image contains lots of edges, in between the ferrite regions, Bainite and Martensite regions, which are dark (blackish) in nature. Ferrite regions are whitish and Bainite and Martensite regions are blackish. The pixels in the edges have similarity with Bainite and Martensite pixels as shown in FIG. 4. To remove the edges of the original images, the image processing technique is applied to the original images. As a result, the phase fraction of the ferrite region is increased and the phase fraction of bainite and martensite regions are decreased. After edge removal, the image with three regions of ferrite, bainite, and martensite are shown in FIG. 6. The edge removal process is illustrated along with FIG. 5 in detail in below. Once edge removed images (101) are received, feature extraction is done along with labeling (102). The following features are extracted from the specific region surrounding each marked pixel, as given below:
[0058] Histogram of contour lines, Histogram of entropy, and a histogram of mean-variance. Labeling of images is done with the help of extracting ground truth images (103). The feature extraction of an image pixel of a marked region is shown in fig-2. After extracting the following features, further concatenating the extracted features are performed to form a feature vector to each marked pixel. Once Labeling is completed, train a random forest (collection of image pixels) with feature vectors and the corresponding labels for phases testing. Trained model (104) is received for further processing.
[0059] Moreover, as per the illustration of FIG. 1b, the test image is taken where the test is to be performed without any marked region. The test image (200) are taken to perform edge removal process (201) and also extracting the features of Histogram of contour lines, Histogram of entropy, and a histogram of mean-variance. After extracting the following features (202), further concatenating the extracted features are performed to form a feature vector corresponding to each pixel of the image. After this process, feed the featured vector to the trained random forest (104). The trained forest provides a class label corresponding to the phase of each pixel. The test image is tested along with the trained model and after testing process output image (203) is received and analysis (204) of output images can be done to know more details on dual phase steel. The phase fraction analysis method is performed.
[0060] In the present subject matter, FIG. 5 illustrates the method of Edge removal process. The process comprises of doing the process of binarizing of the steel original image (100-1) (Ferrite regions, Bainite regions, and Martensite regions) with edges, as a result, binary image of raw steel image (100-2) is received; performing dilation and after removing the boundary pixels of the Bainite and Martensite region, dilated image (100-3) is received; performing the erosion on the dilated image, eroded image (100-4) is received and also performing masking operation on the original image using the binary image; and keeping the black pixel as it is, and replacing the remaining pixels with the while pixels in the original image and getting the edge removed image(101).
[0061] In the present subject matter, FIG. 3 illustrates the method of extracting features as a mean-variance histogram. After calculating the features of each pixel, use the mean-variance histogram of a pixel in a certain neighbourhood. First, a 31x31 neighbourhood around a pixel is taken from the image as shown in FIG. 2. After dividing the 31x31 region into small blocks of size 3x3, For each of the 3x3 blocks (9 pixels), mean and variance is calculated. Let, the lowest mean is µ1 and the highest mean is µ2. Similarly, suppose the lowest variance is s1 and the highest variance is s2. Then, it is divided the range of mean (µ2 - µ1) into 10 equal ranges and also divide the range of variance (s2 - s1) into 10 equal ranges. This way, 100 mean-variance ranges are achieved. Now, for each of the 100 mean-variance ranges, the number of 3x3 blocks whose mean and variance is evaluated and falls into that particular mean-variance range and plot the mean-variance graph (mean-variance histogram) where mean denotes the y-axis, variance denotes the x-axis (i.e, the XY-plane denotes the 100 mean-variance ranges) and the z-axis denotes the number of 3x3 blocks, whose mean and variance belongs to that particular mean-variance range as shown in the FIG. 3. This way, it is received 100 mean-variance bins and their 3x3 block counts. This is represented as a 100 dimensional vector. This is called the feature vector.
[0062] Further, it is classified as the pixels in different phase classes based on the extracted features. For this, it is proposed a novel random forest classifier with an optimal number of trees. This random forest uses automatic feature selection. Classification performance in random forest classifier generally improves with an increase in the number of trees. But experimental evidence have suggested that random forest may overfit data with an excessive number of trees. Further, it has been shown that random forest may perform biased feature selection for individual trees based on the scale of measurement of individual features. As a result, an unimportant feature may be favoured in a noisy feature set. Consequently, classification accuracy may be degraded. So, an increased proportion of important features (removal of unimportant features) may have a significant effect on the classification performance of random forest. So, we introduce an optimal random forest (ORF) which finds the important features (from the feature vector associated with data) and determine an optimal number of trees based on the features to maximize the classification accuracy. The proposed random forest starts with a small number of highly important features (in between a large number of less important features) based on a novel ranking. Then the forest automatically updates the set of important and unimportant features by either finding one new important feature from the set of features or removing at least one unimportant feature depending on certain criteria. The list of important features gets updated with the growth of the forest and the number of unimportant features gets reduced at the same time. The growth of the forest (change in a number of trees) converges producing optimum classification performance. To the best of our knowledge, no other literature proposes a random forest where the number of trees increases automatically leading to monotonic improvement in classification performance and simultaneously reducing unimportant features.
[0063] Again the edge removal image contains whitish ferrite region and blackish/greyish region representing in both bainite and martensite. An example of the edge removed image is shown in FIG. 2 and FIG. 6. The non-white greyish region is subjected to two class classification using random forest-based ensemble classification. A typical steel microscopic image has large portions of ferrite compared to bainite and martensite. When the previous approaches treat classification as a straightforward 3-class problem, there develop class imbalance issues, which is treated in a better way in the proposed hierarchical classification approach. Note that the classification of bainite and martensitic pixels does not have class imbalance problem as bainite and martensite have similar phase fraction values.
[0064] In the present subject matter, once edge removed image is received and features are extracted, a white region of ferrite and blackish regions of (either Bainite or Martensite) are found. Further, it is performed testing only on blackish regions (i.e. Bainite and Martensite pixels) by using k-fold cross-validation process. For example, 5 fold cross-validation process is taken on the non-ferrite regions of the 20 images. At first, the edges of all the 20 images are removed. The concept of k-fold cross-validations is that it is divided the dataset (in this case, the images) into k folds or parts. In each fold, it is performed the testing on the images in one fold and training with the images in the other (k-1) folds. In each fold, we performed training using 16 images and using the trained model, it is performed testing on 4 images.
• In 1st fold, images(1,2,3,4) were used as test images and rest 16 images were used as training images
• In 2nd fold, images(5,6,7,8) were used as test images and rest 16 images were used as training images
• In 3rd fold, images(9,10,11,12) were used as test images and rest 16 images were used as training images
• In 4th fold, images(13,14,15,16) were used as test images and rest 16 images were used as training images
• In 5th fold, images(17,18,19,20) were used as test images and rest 16 images were used as training images
[0065] The phase fractions of the 20 images after 5-fold cross validation are given below in the tabular format.
[0066] Results of phase fraction of three phases are tabulated with the conventional method and present invention method in below:

Nital Image Le Pera Image Algorithm Output Image Phase Fraction (in %)
(Conventional Method) Phase Fraction (in %)
(Present subject matter Method)
Martensite Bainite Ferrite Martensite Bainite Ferrite
1000x_1_With out contour 1000x_1
1000x_1-output 7 16 77 4.76943 16.28939
78.94118

1000x_2_With out contour 1000x_2 1000x_2-output 7 14 78 5.383708
15.40769
79.2086

1000x_3_With out contour 1000x_3 1000x_3-output 7 17 76 5.174276
16.39659
78.42914

1000x_4_With out contour 1000x_4 1000x_4-output 6 19 74 6.451721
15.4541
78.09418

1000x_5_With out contour 1000x_5 1000x_5-output 4 20 76 9.456787
11.65776
78.88546

1000x_6_With out contour 1000x_6 1000x_6-output 4 19 77 9.490723
11.41734
79.09194

1000x_7_With out contour 1000x_7 1000x_7-output 6 16 79 8.206543
12.96149
78.83197

1000x_8_With out contour 1000x_8 1000x_8-output 7 13 80 9.704386
10.90597
79.38965

1000x_9_With out contour 1000x_9 1000x_9-output 6 13 81 7.019328
14.36542
78.61525

1000x_10_With out contour 1000x_10 1000x_10-output 6 14 80 6.521749
14.42869
79.04956

1000x_11_With out contour 1000x_11 1000x_11-output 3 17 80 4.475328
11.80245
83.72222

1000x_12_With out contour 1000x_12 1000x_12-output 4 20 77 5.394706
11.09123
83.51406

1000x_13_With out contour 1000x_13 1000x_13-output 5 17 78 4.095456
12.583
83.32154

1000x_14_With out contour 1000x_14 1000x_14-output 4 16 80 4.11858
15.71196
80.16946

1000x_15_With out contour 1000x_15 1000x_15-output 6 14 80 7.723876
10.07859
82.19754

1000x_16_With out contour 1000x_16 1000x_16-output 5.204214
12.36684
82.42894

1000x_17_With out contour 1000x_17 1000x_17-output 3 17 80 3.614919
12.01497
84.37011

1000x_18_With out contour 1000x_18 1000x_18-output 5 15 80 5.852083
11.29777
82.85015

1000x_19_With out contour 1000x_19 1000x_19-output 5 18 77 6.186307
10.12043
83.69326

1000x_20_With out contour 1000x_20 1000x_20-output 4.015362 11.66888
84.31576

[0067] The nital etched and La-Pera etched microscopic images of dual phase steel and its microstructure are used as the database for storing million of pixel samples. In total, more than a million (1357086) pixel samples from bainite, martensite, and ferrite classes are used for experiments. Out of these samples, 162672 samples are bainite, 385036 samples are martensite and 809378 samples are ferrite.
[0068] In the present subject matter, it is introduced a hierarchical classification scheme. In the hierarchical model, first, ferrite, the dominant class is detected. Second, in the remaining region of the image, bainite and martensite are estimated. Further, the nital-etched image pixel values (multi-valued function) have variation in intensity. These images are binarized (black and white images) using an image processing technique. This is followed by repetitive cycles of dilation and erosion, an image morphological operation. This process removes edges of the nital-etched image. In the present subject matter, it is approached the domain of estimation of phase fraction, based on the edge removal process. The edge removed image contains a whitish ferrite region and greyish region representing both bainite and martensite. An example of the edge removed image is shown in FIG. 2. The non-white greyish region is subjected to two class classification using random forest-based ensemble classification. A typical steel microscopic image has large portions of ferrite compared to bainite and martensite. It is observed that the classification of bainite and martensitic pixels does not have class imbalance problem as bainite and martensite have similar phase fraction values.
[0069] One of the specific features of the present disclosure is to objectively enhance the discriminatory power of features. Mean and variance of pixel values around an image location (gray pixels of an edge removed the microscopic image) are taken as features. However, the straightforward use of these features does not discriminate bainite and martensite pixels significantly as shown in Fig. 7. FIG. 7 is illustrated the mean-variance feature plot where X-axis is mean and Y-axis is variance and redpoint denotes martensite and green point denotes bainite pixels. This is shown also in FIG. 4, where bainite and martensite have significant class overlap. In order to reduce the feature space overlap, we calculate the multilevel mean and multilevel standard deviation for each pixel for training and testing nital-etched image segmentation as a two-class classification problem. The definition of multilevel mean and multilevel standard deviation is disclosed separately. Overall, the idea of multilevel feature computation at a particular image location is to give higher weights of local or closer pixels to the central pixel in question for which features are being calculated. The multilevel mean and multilevel standard deviation of the same two classes as shown in Fig. 3 is plotted in Fig. 8(a). Fig. 8(b) and 8(c), respectively show feature distributions of martensite and bainite classes separately. This is an important contribution and to the best of our knowledge, previous phase fraction analysis methods from microscopic images did not show any such objective feature analysis.
[0070] FIG. 8A is illustrated in the multilevel mean–standard deviation plot where X-axis is multilevel-mean and Y-axis is a multilevel-standard deviation and red point represents martensite and green point represents bainite pixels.
[0071] FIG 8B is illustrated in the multilevel mean–standard deviation feature plot for Bainite class only where X-axis is multilevel-mean and Y-axis is a multilevel-standard deviation and green point represents bainite pixels.
[0072] FIG. 8C is illustrated in the multilevel mean–standard deviation plot for martensite class only where X-axis is multilevel-mean and Y-axis is a multilevel-standard deviation and red point represents martensite.
[0073] Segmented image after classifying martensite (red pixels) and bainite (green pixels) classes in a ferrite base (blue pixels) for different features are shown in FIG. 9A and FIG. 9B.
[0074] In the next step, we are exploring a distance measure that can identify outliers within multilevel mean standard deviation feature space.
[0075] Multilevel mean and standard deviation: By the term ‘Multilevel’, it is meant that the processes are using the neighbourhood information of a pixel on the basis of locality (i.e., when we are calculating the feature of a pixel, the nearest pixels will contribute more to the feature, than the furthest pixels) within a certain neighbourhood. In the present subject matter application, it is used 41x41 neighbourhoods.
[0076] It is calculated Mean and Standard Deviation for each of the neighborhoods i, where, i=1,2,…,20 (as shown in Fig 6). Then it is added all the Mean or Standard Deviation values and get Multilevel Mean or Standard Deviation. The black dot in Fig-2 is the pixel for which we are calculating features. Fig 10 is illustrated the description of Multilevel Mean and Standard Deviation Calculation based on below relations.
Multilevel Mean = ?Mean (pixel values within neighborhood i), where i = 1,2,…,20
Multilevel Std = ?Standard Dev (pixel values within neighborhood i ), where i = 1,2,…,20
[0077] Furthermore, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. It will be appreciated that several of the above-disclosed and other features and functions, or alternatives thereof, may be combined into other systems or applications. Various presently unforeseen or unanticipated alternatives, modifications, variations, or improvements therein may subsequently be made by those skilled in the art without departing from the scope of the present disclosure as encompassed by the following claims.
[0078] The claims, as originally presented and as they may be amended, encompass variations, alternatives, modifications, improvements, equivalents, and substantial equivalents of the embodiments and teachings disclosed herein, including those that are presently unforeseen or unappreciated, and that, for example, may arise from applicants/patentees and others.
[0079] While the foregoing describes various embodiments of the invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof. The scope of the invention is determined by the claims that follow. The invention is not limited to the described embodiments, versions or examples, which are included to enable a person having ordinary skill in the art to make and use the invention when combined with information and knowledge available to the person having ordinary skill in the art.