WO2015032835A1

WO2015032835A1 - Method and system for detection of fluid inclusion

Info

Publication number: WO2015032835A1
Application number: PCT/EP2014/068765
Authority: WO
Inventors: Martin VAD BENNETZEN
Original assignee: Mærsk Olie Og Gas A/S
Priority date: 2013-09-03
Filing date: 2014-09-03
Publication date: 2015-03-12
Also published as: DK201570249A1

Abstract

A method and a system for automatic detection of fluid inclusions in crystalline materials,such as rocks, are provided. The method comprises receiving at least one digital image of a crystalline material, determining global image intensity properties of each pixel of the received digital image and applying one or more global image filters on the determined global image intensity properties to provide a first filtered image. A set of filters are successively applied by segmenting the first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections to provide a further filtered image. A resulting filtered image is provided and based on the resulting filtered image, fluid inclusions in the at least one received image are identified.

Description

METHOD AND SYSTEM FOR DETECTION OF FLUID INCLUSION

TECHNICAL FIELD The invention relates to a method and a system for detection of fluid inclusions, such as detection of fluid inclusions in rock formations, and in particular to automated detection of fluid inclusions using image analysis, such as using intensity analysis of images which may contain fluid inclusions. BACKGROUND OF THE INVENTION

Especially in the mining industry and the oil and gas industry, the

understanding of the geological environment in a given area is valuable, and getting an understanding of rock properties during diagenesis allows for understanding of the geological history which in turns provides for estimating the potential for rock reservoir development.

Fluid inclusions are microscopic bubbles of a fluid, such as a liquid, typically water or petroleum, or a gas, which is trapped within a crystal. The trapped fluids inside an inclusion preserve information about the physico-chemical conditions prevalent during diagenetic and crystal growth. Thus, from analysing the fluid trapped in the fluid inclusions, it is possible to determine various reservoir properties at the time of diagenesis, such as chemical composition of the fluid, temperature and pressure.

Typically, when analysing images of rocks which may contain fluid inclusions, the analysis of the images are performed manually by inspection and thus may provide for a subjective quantification of the images and possible inclusions. In metallurgy, inclusions such as microscopic particles of a chemical composition different from a composition of the metal alloy, may be quantified by inspection, see for example US 8,347,745. Furthermore, single-image spot detectors are known. Presently, the focus in the development of spot detectors is on bioscience research and the detectors are typically optimized to detect spots, such as particles or cells, being well-defined and typically marked with bio-markers to obtain good contrast properties, for example using 2D gel electrophoresis autographs involving radioactivity, DNA arrays involving fluorescence, etc.

However, even though fluid inclusions may be detected with the above image analysis methods, none of the proposed detectors allow for automated detection of fluid inclusions.

SUMMARY

It is an object of the present invention to provide an improved method and system for detection of fluid inclusions.

According to one aspect of the present invention, a method of automatic detection of fluid inclusions in crystalline material is provided. The method comprises receiving at least one digital image of a crystalline material, determining global image intensity properties of each pixel of the received digital image and applying one or more global image filters on the determined global image intensity properties to provide a first filtered image. A set of filters may be applied successively by segmenting the first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections, and providing a further filtered image. A resulting filtered image may be provided after applying one or more of the set of filters, and based on the resulting filtered image, fluid inclusions in the at least one received image may be identified. According to one aspect of the present invention, a method of automatic detection of fluid inclusions in crystalline material is provided. The method comprises receiving at least one digital image of a crystalline material, determining image intensity properties of each pixel of the received digital image and applying one or more global image filters on the determined image intensity properties to provide a first filtered image. A set of filters may be applied successively by segmenting the first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections, and providing a further filtered image. A resulting filtered image may be provided after applying one or more of the set of filters, and based on the resulting filtered image, fluid inclusions in the at least one received image may be identified. Segmenting may comprise or mean performing a division of an element into a number of segments, such as into smaller segments, i.e. dividing the image into a number of segments or parts.

The term global image intensity properties may be termed image intensity properties, likewise the global image intensity properties may comprise image intensity properties of each pixel in the received digital image. The term global may thus refer to the entire received digital image, such as to the image before segmentation is performed, and for example global properties may define properties of the entire image and likewise global filters may define filters filtering the entire image, etc.

The method may be implemented in a computer, and the at least one digital image may be received at a processor configured to determine global intensity properties, applying one or more global filters. The same or one or more further processors may be provided and may be configured to successively apply one or more filters selected from a set of filters. According to a further aspect of the invention a method of automatic detection of fluid inclusions in crystalline materials, is provided, the method comprising receiving at least one digital image of a crystalline material and successively applying a set of filters by segmenting a first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections, to provide a further filtered image, and providing a resulting filtered image. Based on the resulting filtered image, fluid inclusions in the at least one received image may be identified.

According to a further aspect of the present invention, a system for fluid inclusion detection, such as a system for automatic detection of fluid inclusions in crystalline materials, is provided. The system comprises a processor configured to: receive at least one digital image of a crystalline material, determine global image intensity properties of the received digital image and applying one or more global image filtering criteria on the determined global image intensity properties to provide a first filtered image. The system may further comprise a storage, such as an electronic storage, for storing the at least one digital image and at least temporarily the first and further filtered images. The processor may be further configured to

successively apply a set of filters by segmenting the first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections, and providing a further filtered image. The processor may be configured to provide a resulting filtered image, and based on the resulting filtered image, identifying fluid inclusions in the at least one received image.

The processor may be configured to provide positions for the identified fluid inclusions. Preferably, the system comprises a user interface, wherein for example predetermined parameters may be set by a user, and furthermore, the processing of a single image or a batch of images may be selected.

The system may further comprise an interface for providing the fluid inclusion positions to a camera, a microscope or an analyser for facilitating further investigation of the identified fluid inclusions.

In a further aspect of the invention a computer program comprising program code means for performing the steps of the disclosed method when said computer program is run on a computer is provided.

In a still further aspect of the present invention, a computer readable medium having stored thereon program code means for performing the method as herein disclosed when said program code means is run on a computer is provided.

It is an advantage of the present invention that fluid inclusions may be identified automatically. This allows for large-scale image analysis and thus multiple images may be analysed within a short time-frame. It is a further advantage that the present invention allows for detection of fluid inclusions even in images with noisy regions, such as in images with regions with very noisy regions.

The received digital image may be a bitmap image and may typically comprise a plurality of pixels, each pixel having one or more intensity values associated therewith. Thus, the image may be represented by pixels and associated intensity value or values. The pixel may be characterised by pixel coordinates in the received image. The image intensity properties may be the intensity value(s) for each pixel, or the image intensity properties may be a function of the intensity values, such as an intensity distribution for the image. The received digital image may be a colour image, and the received digital image may comprise pixel and colour intensity values, such as intensity values in a number of channels, typically such as intensity values for a red channel, a blue channel and a green channel. In one or more embodiments, a grey-scale digital image may be provided by adjusting intensity values of the received digital image using a linear correlation between intensities of the number of channels, such as between intensity values of a red, a green and a blue channel in the received image to provide a corresponding grey-scale intensity image. It is an advantage that a single intensity value may be provided for each pixel in a grey-scale image. The grey-scale image may be any grey-scale image, such as an 8-bit grey-scale image, or a 16-bit grey- scale image. An image may be represented by a dataset of pixel and corresponding intensity value(s). The digital image may be an optical image, such as a light microscope image, the light microscope image being obtained by using visible light.

The at least one digital image may be stored in a storage, such as an electronic storage, such as a digital library. The storage may form part of the system, such as the system for processing the images, or the storage may be provided externally of the system and for example be accessible by the processor. The digital library may contain any number of images to be analysed, such as between 1 and 10, such as more than 10, such as more than 100, such as more than 1000, such as more than 100,000. The digital library may contain complete image files in any format, and/or the digital library may comprise datasets comprising pixel and corresponding intensity value(s) representing an image.

The material may be any material, such as minerals, rocks, crystalline materials, crystalline minerals, etc. An image of the material may be acquired by any camera, such as by any microscope, such as a light microscope, an interference microscope, a differential interference contrast (DIC)

microscope, a Nomarski microscope, a phase contrast microscope, an electron microscope, etc. Typically, the material is prepared as thin-sections, and an image of the thin-section is acquired by the microscope.

In one or more embodiments, the microscope may be a robotic microscope capable of scanning the material, such as the thin section material, while acquiring images. The images may be received, e.g. by a processor, and may be analysed in real-time. This is especially advantageous for detection of fluid inclusions in minerals, as there may be a number of samples or thin sections in which no fluid inclusions are present. By being able to analyse the images in real-time, a decision of whether to provide the sample or thin- section for further analysis or not may be taken efficiently and during scanning, eliminating waiting time, and cumbersome manual analysis of hundreds of images before allocating a sample or thin section for fluid inclusion analysis.

In one or more embodiments, position information for the identified fluid inclusions may be provided by the method, for example by providing coordinates for the identified fluid inclusions, or by marking an area of the sample or thin section comprising the identified fluid inclusions. The position information may be provided to a fluid inclusion analyser for automatic positioning or zooming in onto the fluid inclusions by the analyser. Typically, fluid inclusions are of a size between 0.1 μιτι and 20 μιτι, such as between 2 μιτι and 15 μιτι, such as between 2 μιτι and 7 μιτι.

A fluid inclusion may be a cavity in a rock formation or in a crystalline material, such as a fluid-filled cavity in a rock formation or crystalline material. Detection of fluid inclusions includes detection of the actual inclusion, such as of the entire fluid inclusion, and may include detection of intensity distribution(s) characteristic for fluid inclusions at and immediately around fluid inclusions.. The detection of fluid inclusions may comprise detection of image properties of the fluid inclusion, such as image properties of the fluid and/or image properties of at least a part of the rock formation or the crystalline material appearing to be a fluid inclusion. The image properties characteristic for fluid inclusions does not necessarily include discontinuities typically used for edge detections.

In one or more embodiments, global image intensity properties are

determined, for example by determining global image intensity values of each pixel of the received digital image and determining a global intensity distribution for the received digital image. Based on the global image intensity properties, one or more global image filters may be applied to provide a first filtered image. The global image intensity properties may be the intensity value(s) for each pixel, or the global image intensity properties may be a function of the intensity values, such as a global intensity distribution for the image.

Alternatively and/or additionally, global image intensity properties may be termed image intentisty properties, thus in one or more embodiments, image intensity properties are determined, for example by determining image intensity values of each pixel of the received digital image and determining an intensity distribution or a global intensity distribution for the received digital image. Based on the image intensity properties, one or more image filters or global image filters may be applied to provide a first filtered image. The image intensity properties may be the intensity value(s) for each pixel, or the image intensity properties may be a function of the intensity values, such as an intensity distribution or a global intensity distribution for the image. In one or more embodiments, fluid inclusions may be identified in an image by successively determining section-specific intensity properties for a plurality of sections of the received image and successively applying section-specific intensity filters in dependence on the determined segment-specific intensity properties. An image may be divided or segmented into a number of sections to analyse the intensity within each section independently. The sections are typically distinct, however, they may overlap with a few number of pixels, such as 5 or 10, to ensure processing of the entire image. The sections may have any form or shape, and may thus be rectangular sections, or circular sections, such as concentric circular sections, etc. Typically, the sections are rectangular of the type n x m pixels, where m may be equal to n to provide a square.

Segmenting the image may comprise or may mean dividing the image into a number of segments or parts.

Fluid inclusions have been carefully studied by the present inventor and it has been found that the intensity distribution at and immediately around fluid inclusions have certain characterising features; for example, the intensity of the fluid inclusion itself is typically characterised by low-intensity pixels, while the fluid inclusion may be surrounded by high-intensity pixels. Other features typically found in the images and more or less obscuring the fluid inclusions are blue-staining, noise, fractures, etc.

The intensity of an image may be roughly indicated as being "low", "medium" or "high", referring to the intensity values of the image. A low intensity may be characterised as intensities having intensity values in the lower four tenths of an intensity spectrum for the image, a medium intensity may be

characterised as having intensity values in a centre of the intensity spectrum, such as between four tenths and seven tenths of the intensity spectrum, and high intensity may be characterised as having intensities above seven tenths of the intensity spectrum. Thus, for intensity values distributed from 0-255, low intensity may be between intensity values of 0-100, such as from 25-100, such as from 50-100, medium intensity may be between 100 and 175, and high intensity may be for intensity values larger than 175, and

correspondingly for any other distribution of intensity values. A very low intensity may be an intensity in the first tenth, and a very high intensity may be an intensity higher than nine tenths of the intensity spectrum.

It has been found that blue-staining which typically stems from epoxy, such as from epoxy glue made to prepare the samples, and noise have

corresponding intensities, and are typically found having a medium intensity. Furthermore, the density of pixels having this intensity has been found to be significantly higher relative to other regions of the image.

It has therefore been found that it is advantageously for the method to further comprise the step of segmenting the received digital image into a plurality of initial sections, determining a sum of intensity values for each of the plurality of initial sections, applying an initial filter on the determined sum of intensity values for each of the plurality of initial sections, and providing an initially filtered received image. The initially filtered received image may be provided to the processor, and the processor may deem the initially filtered received image, the received image for processing.

The set of filters may comprise any number of filters suitable for filtering of the received image. The set of filters may comprise first, second, third and fourth filters. In some embodiments the set of filters may comprise the initial filter and the global filter.

The initial filter may comprise determining a sum of pixel intensities within each of the initial sections, and for each initial section wherein the sum of pixel intensities is less than an initial threshold intensity value, setting all pixel intensities in that initial section to a predeternnined value to thereby filter out blue staining and/or noise.

- 7-' 1 *p^si-x < I ^lmax > wherein Ip_ix is the pixel intensities in the section S, and Lax is the initial threshold intensity value.

The initial sections may be rectangular sections, such as sections comprising n x m pixels, n may be set equal to m. The segmentation of the received digital image into a plurality of initial sections may be predetermined. Thus, the segmentation may be provided independently of the determined intensity properties.

The initial threshold intensity value may be determined as a function of the size of the sections, and is typically a very high value, in order to exclude only noise and blue-staining and without excluding possible fluid inclusions. In one example, the section is 100 x 100 pixels, the intensity values are distributed between 0 and 255, and the maximum intensity value, i.e. the initial threshold intensity value is 1 ,700,000.

Throughout the description, typically, the predetermined value is mentioned as being zero, and the value zero has been mentioned throughout the description, however, it is envisaged that the pixels to be filtered out may be set to any predetermined value and the filtering out may be performed mathematically, for example, the predetermined value may include a very low intensity, or a very high intensity, such as the highest intensity value.

In one or more embodiments, the segmenting of the filtered image and/or of the received image may be predetermined. Thus, the segmentation may be provided independently of the determined intensity properties. Furthermore, the size of the plurality of sections may be predetermined, i.e. m and n may be predetermined for each filter. The global image filter may be applied by calculating a global intensity distribution of the received image, setting a global pixel threshold intensity based on the global intensity distribution of the received image, and for all intensities of the received image being greater than the global pixel threshold intensity, setting the intensity to zero to thereby provide a globally filtered image.

It is an advantage of using a global image filter in which the global intensity distribution of the received image, i.e. the intensity distribution of the entire image including all pixels, is taken account of when setting the global pixel threshold intensity. In this way, the method of identifying fluid inclusions may be reliable also when there are inter-image variations, such as intensity variations among different images.

In one or more embodiments, the global pixel threshold intensity may be determined as a function of the global intensity distribution, and the global pixel threshold intensity may for example be determined as the p^th-quantile of the intensity distribution of the received image, p may be any number below 0.05, such as below 0.01 , such as below 0.005, such as 0.005, or such as 0.002.

In one or more embodiments, the set of filters may comprise a filter in which the received image, which may be further filtered, is segmented into a number of predetermined sections, such as into sections of a size n x m, wherein n and m may be predetermined for the specific filter. It is determined that if a number of pixels having non-zero intensities within a section is larger than a threshold, then all pixel intensities in that section is set to a

predetermined value, such as zero. Thereby a further filtered image is provided. The threshold may be multiplied with a scaling-factor g related to conditional handling of very large fluid inclusions to account for for example large fluid inclusions which are typically less intense than "noisy" regions.

In one or more embodiment, the set of filters may comprise a filter in which the received image, which may be further filtered, is segmented into a number of predetermined sections, such as into sections of a size n x m, wherein n and m may be predetermined for the specific filter. It is determined that if a number of pixels having non-zero intensities within a section is less than a threshold, then all pixel intensities in that section is set to a

predetermined value, such as zero. Thereby a further filtered image is provided.

The set of filters may comprise any number of filters, such as a plurality of filters, such as two or more filters, etc. The set of filters may comprise a first filter, a second filter, a third filter, a fourth filter, etc. and a predetermined part of the set of filters may be applied, or all the filters in the set of filters may be applied.

In one or more embodiments, a first filter using a number of pixels having non-zero intensities within a section may be implemented, the threshold being a first threshold. The filtered image may be segmented into vertical and/or horizontal slice-formed sections, each slice-formed section having a predetermined first size. Generally, the slice-formed sections may be sections n x m, or vice versa, thus the slice-formed sections may be bands or strips, and they may be provided horizontally, vertically or cross wise at the image. A slice-formed section may in some embodiments have a length along an entire width of the image and a predetermined width, i.e. the first size, or the slice-formed sections may be sections having a length along an entire height of the image and a predetermined width, i.e. a predetermined first size. Thus, m, n may be any value from between 1 to the maximum number of pixels in the

corresponding direction, and in some embodiments m may be two, five or ten times higher than n. In some embodiments, the filter may employ horizontal slice-formed sections and vertical slice-formed sections successively, and the predetermined width of the vertical slice-formed sections may be different from the width of the horizontal slice-formed sections.

By filtering using horizontal or vertical slice-formed sections as mentioned above fractures in the rock and/or noise may be eliminated from the image. The fractures eliminated may be rough fractures.

The set of filters may comprise a second filter which may be implemented with a second threshold. Using the second filter, then if a number of pixels having non-zero intensities within a section is less than the second threshold, then all pixel intensities in that section are set to a predetermined value, such as zero. The filtered or further filtered image is segmented into rectangular sections of a predetermined second size of n x m.

The second threshold is set so that areas or sections in which only a few pixels have intensities larger than zero, then these areas or sections are filtered out, as it is assumed that if only a few pixels have intensities which are non-zero, then those pixels do not form part of a fluid inclusion.

In one or more embodiments, the set of filters may comprise a third filter, wherein the filtered image is segmented into vertical and/or horizontal slice- formed sections of a predetermined third size, and wherein for each slice- formed section in which a sum of pixel intensities is larger than a third threshold, then the pixel intensity in that section is set to zero. Typically, the third size will be smaller than the first size, and thus the slices or bands used with the third filter may be thinner than the slice-formed sections used with the first filter. Typically, the first filter will be used with either the horizontal si ice-formed sections or vertical slice-formed sections, whereas the third filter may typically be used with horizontal and vertical slice-formed sections sequentially. It is an advantage of the third filter that remaining fractures, which were not eliminated by using the first filter, may be removed.

The set of filters may further comprise a fourth filter; the filtered image may be segmented into rectangular sections, and wherein for each rectangular section in which a sum of pixel intensities is less than a fourth threshold, then the pixel intensity in that section is set to zero. The fourth filter may be a post- processing filter, and typically, the fourth threshold is determined so that some false positive fluid inclusions may be removed.

From the resulting image, provided after one or more of the filters from the set of filters have been applied, a resulting filtered image is provided in which fluid inclusions may be identified. Generally, all spots left in the resulting filtered image may be fluid inclusions and may be identified as such. The positions of the identified fluid inclusions in the at least one digital image may be provided, for example as a control signal to a microscope, a fluid inclusion analyser, a camera, etc.

Thus, the method may further comprise the step of analysing an identified or validated fluid inclusion.

The identified fluid inclusions may be validated if a number of validation criteria are fulfilled. The validation criteria may be based on the received digital image, or they may be based on the resulting filtered image. The fluid inclusions may be validated if one or more of the following validation criteria are fulfilled: a) the resulting filtered image has a summed intensity /_ioi._res of at least l_min;

I tot. res — I min b) a standard deviation of a global intensity distribution of the received digital image <¾ is below a threshold intensity distribution a_max.init_;

c) the mean value μ^ of a global intensity distribution in the resulting filtered image is less than the mean value μ_ηΛ of the global intensity distribution in the received digital image at a significance level of a, so that a p-value, p, is less than a;

p< a

with a significance level of a, for example using the one-sided test for down shift of mean Welch T-test, a may be equal to 0.0001 . d) the mean-shift from the mean value μ,ηκ θί the global intensity

distribution in the received digital image to the mean value of the intensity distribution of the resulting filtered image jL futer is greater than β times the standard deviation of the global intensity distribution of the received image, in one example β may equal 2. μ,ηΚ - ^filter > β Jinit e) The noise-to-signal ratio of the receive digital image must be below a critical ratio C, C may be equal to 4, (e.g.: C=0.4), defined as the ratio between the standard deviation of the global intensity distribution of the received digital image <¾ and the mean value μ_ηΛ of of the global intensity distribution in the received digital image where C is between 0 and 1 , f) a score S is greater than a critical score S_c, such as S_c = 8, wherein S may be defined as:

and in one or more embodiments, S may be defined as:

^ ^init

wherein the weighting factors Ki, K₂, K₃, K4 and K₅ are real numbers. In one or more specific examples Ki - K5 may be 1 , 3, 1 , -3 and 30 respectively,

-log (p) is a measure of the magnitude of the mean-shift

•

is a measure of how much smaller the noise-to-signal ratio is compared to the pre-set critical ratio · (? L \ j_{s a} measure of the initial signal-to-noise ratio

• (?ML \ j_{s a} measure of the noise-to-signal ratio,

• ^{hot res} - 1 is a measure of how different the initial and final post-processed image are.

Thus, after validation of the images, the fluid inclusions are validated fluid inclusions, and any further processing may be performed on the basis of the validated fluid inclusions only. The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the invention are shown.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and other features and advantages of the present invention will become readily apparent to those skilled in the art by the following detailed description of exemplary embodiments thereof with reference to the attached drawings, in which: Fig. 1 illustrates an exemplary image for analysis,

Fig. 2 illustrates a flow diagram of a method for detection of fluid inclusions,

Fig. 3 illustrates an exemplary image after applying a step of the method,

Fig. 4 illustrates an exemplary intensity distribution of an exemplary image for analysis Fig. 5 schematically illustrates a step of the method comprising sectioning the image,

Fig. 6 illustrates an exemplary image after applying a initial filter of the

method,

Fig. 7 illustrates an exemplary image after applying a global filter of the

method,

Fig. 8 schematically illustrates a step of the method comprising vertically slicing the image,

Fig. 9 schematically illustrates a step of the method comprising horizontally slicing the image, Fig. 10 illustrates an exemplary image after applying the method,

Fig. 1 1 illustrates a region of an image containing a number of potential fluid inclusions, Fig. 12 illustrates a flow diagram of a method for validating an automatic detection of fluid inclusions,

Fig. 13 illustrates a flow diagram of a method for detection of fluid inclusions, incorporating detection and validation, Fig. 14 schematically illustrates an exemplary system for detection of fluid inclusions.

DETAILED DESCRIPTION

The figures are schematic and may be simplified for clarity, and they merely show details which are pertinent to the understanding of the invention, while other details have been left out. Throughout, the same reference numerals are used for identical or corresponding parts.

Figs. 1 a and 1 b illustrate an exemplary image 2 of a rock sample, such as a thin section of a rock prepared for analysis. Fig.1 a is a colour image, and Fig. 1 b is the colour image in a black and white version Typically, when analysing such rock samples, fluid inclusions are of special interest. Fluid inclusions are microscopic bubbles of a fluid, such as a liquid, typically water or petroleum, or a gas, which is trapped within a crystal. The fluid inclusions are typically between 0.1 and 20 μιτι and may not have a predetermined shape or form and it is therefore difficult to identify the fluid inclusions using standard imaging analysis.

The order of the steps of the method as disclosed below may be maintained, typically, this will provide the best results, or the steps of the method may be performed in a different order.

The image 2 is in this example a colour image provided from a light microscopy, however the image may also be provided from other sources. The image 2 may be a bitmap image with any resolution, such as a resolution of e.g. 2560 x 1920. In some embodiments, the image may originate as a grey-scale image. The image 2 shows a region 4 of fluid inclusions 10, the region 4 being marked by a square for illustrative purposes.

Typically, each pixel in an image may be represented by intensity, such as by an intensity value, such as represented by an integer, such as by an integer between 0 and 255 for an 8-bit image. In a colour image each pixel may be represented by three intensities in a red channel, a green channel and a blue channel, respectively.

Fig. 2 illustrates a flow diagram of a method 100 of automatic detection of fluid inclusions in crystalline materials. The detection method 100 comprises an optional first step of applying a conversion 102 provided to convert a received colour image comprising three intensity values for each pixel to an image comprising only a single intensity value for each pixel, such as a grey-scale image. In the step 104, global image intensity properties of each pixel of the received digital image is determined, and in step 106, one or more global image filters on the determined global image intensity properties are applied to provide a first filtered image. In steps 108-1 12 a set of filters are successively applied: in step 108, the first or a further filtered image is segmented into a plurality of sections, in step 1 10 a filter from a set of filters is applied on one or more determined intensity properties for each of the plurality of sections, and in step 1 12 a further filtered image is provided, until a predetermined number of filters from the set of filters have been applied, and a resulting filtered image is provided in step 1 14. In step 1 16, based on the resulting filtered image, fluid inclusions are identified in the at least one received image. The result of applying the conversion 102 is illustrated in Fig. 3, wherein an exemplary grey-scale image 6 corresponds to the image 2 of Fig. 1 after applying the conversion 102.

The purpose of the conversion 102 is to provide an image with a single intensity value per pixel. In cases where the image is provided as an image with only a single intensity per pixel, i.e. a grey-scale image, the conversion 102 may be omitted. Furthermore, it is envisaged that the method may be applied using a colour image having more than one intensity value per pixel, however, the processing power and the computational complexity may be significantly increased.

To convert a colour image 2 to a single intensity image, i.e. a grey-scale image, 6, the conversion 102 may apply the following formula:

Igrey ~ d Ired b Igreen C lblue>

wherein the intensity / of each of the colours is weighted by a, b and c, and summed to form a combined intensity l_grey. To provide /_grey on the same scale, e.g. 0-255, as the individual colour intensities, the factors a, b and c should sum to 1 . In an exemplary method, a may be in the range of 0.15-0.45 such as in the range of 0.25-0.35, b may be in the range of 0.45-0.75 such as in the range of 0.55-0.65 and c may be in the range of 0.01 -0.25 such as in the range of 0.05-0.15. The values of a, b and c may be fine tuned from empirical analysis. In one specific example, a general formula is used, wherein a=0.2989; b=0.5866 and c=0.1 145.

By studying a large amount of intensity maps of rock images, the present inventor has succeeded in quantifying different areas of the intensity map as stemming from different features.

It has for example been found that large areas of pixels having medium intensities, such as intensities between 100 and 175, such as area 8 as marked in Fig. 3, are typically noisy regions or intensity stemming from blue staining from epoxy glue with which the rock samples are prepared. The fluid inclusions 10 are typically low intensity pixels, i.e. intensity below 100, surrounded by areas of high intensity pixels, i.e. pixels having intensities above 175.

Fig. 3a shows the resulting grey-scale image as visualized by a heat map, including a colour code for the intensity values so that green colours correspond to high intensities, yellow colour to medium intensities and red colours to low intensities. Fig. 3b shows the heat map in black and white, thus dark green colours, corresponding to a high intensity and dark red colours, corresponding to a low intensity may both be seen as dark colours.

In Fig. 4, an intensity distribution 30, and the colour key, of the exemplary image 6 is shown. The 1 ^st axis 32 denotes the intensity from 0 to 255, and the 2^nd axis denotes the count of pixels with a particular intensity. Low intensity is defined as the lower part of the spectrum 36, medium intensity as the middle part of the spectrum 38, and high intensity as the higher end of the spectrum 40. The detection method 100 may comprise an initial filter 104 to disregard parts of the image having a high density of medium intensity pixels. Applying the initial filter 104 comprises segmenting the image into sections, as illustrated in Fig. 5 which shows sectioning a schematic image 50 in sections 52 of n x m pixels 54, in this particular example n = m hence, the sections are quadratic sections 52. The size of n may be chosen on basis of the resolution of the image 50, n may in an exemplary method, be selected between 25- 200, such as between 50-150, such as between 75-125.

Within each section S 54, the intensity values of the pixels l_Pi_X may be summed and a criterion may be applied on the sum of pixel intensities such that if the criterion is fulfilled, intensity values of all pixels within the section 54 is excluded e.g. set to 0. In the exemplary detection method 100, the criterion of the initial filter 104 may be that the summed intensities of pixels within the section is below an initial threshold intensity value, i.e.:

The initial threshold intensity value, i.e. I_max used in the initial filter 104 may be dependent on the size of the section, and the overall intensity of received image. An exemplary image 12 is shown in Fig. 6 illustrating the initially filtered received image 12 as a result of applying the initial filter 104 to the image 6. It is seen that the noisy and blue stained regions 8 have been excluded by setting the intensity values of those pixels to 0.

Fig. 6a shows the filtered image as visualized by a heat map, including a colour code for the intensity values so that green colours correspond to high intensities, yellow colour to medium intensities and red colours to low intensities. Fig. 6b shows the heat map in black and white, thus dark green colours, corresponding to a high intensity within areas of lower intensity are seen as darker spots on a lighter background, whereas a solid dark red colours, corresponding to a low intensity is seen as a solid dark colour.

As noted, fluid inclusions predominantly contain pixels with lower intensities, such as intensities below 100. Moreover from the intensity distribution 30 (Fig. 4) showing the intensity values of the converted image 6 (Fig. 3), it may be seen that intensities below 100 are in the lower tail of the distribution 30. Therefore, a global filter 106, 108 is applied. In 106, global intensity properties, such as the intensity distribution, for the global image are determined. The global filter uses the intensity distribution 30 of the

converted image 6, to exclude every pixel with an intensity greater than the p^th quantile of the intensity distribution 30. Thus, the pixels to be excluded or filtered out in the global filter 106, 108 are the pixels satisfying:

Ipix > ^p^thquantile - where l_Pi_X is the pixel intensity value, and I_pth_quantile is the global pixel threshold intensity. The pixels may be filtered out by setting their intensity values to 0. p may be assigned a value less than 0.01 , such as between 0.001 and 0.008, such as between 0.005 and 0.008, or such as between 0.001 and 0.003, such as 0.002.

The global filter 106 may be applied to the result of the initial filter 104. Fig. 7 shows a globally filtered image 14 showing the result of applying the global filter 106, 108 to the initially filtered received image 12, wherein the globally filtered image 14 shows that a majority of pixels have been excluded due to the application of the global filter 106. However, it is seen that the fluid inclusions 10 are not excluded. There are however, still scattered pixels 16 that are not excluded, and which are not fluid inclusions.

Fig. 7a shows the filtered image as visualized by a heat map, including a colour code for the intensity values so that green colours correspond to high intensities, yellow colour to medium intensities and red colours to low intensities. Fig. 7b shows the heat map in black and white, thus dark green colours, corresponding to a high intensity within areas of lower intensity are seen as lighter spots, whereas a solid dark red colours, corresponding to a low intensity is seen as a solid dark colour.

The set of filters 1 10 may comprise a number of filters, 109, 1 1 1 , 1 13, 1 15 and the filters 109, 1 1 1 , 1 13, 1 15 may be applied successively, either in the order as described or in any other order. Either the entire set of filters 1 10, including first filter 109, second filter 1 1 1 , third filter 1 13 and forth filter 1 15 may be applied or any part of the set of filters 1 10 may be applied.

Applying the first filter 109 comprises segmenting the globally filtered image as illustrated in Fig. 8, wherein the image 50 is segmented into vertical slices 56, or vertical slice-formed sections 56, 58. Each vertical slice 56, 58 may have a predetermined width of n pixels, wherein n may be between 50 and 150, such as between 75 and 125 such as 100. Alternatively or additionally, applying the first filter 109 may comprise segmenting the image 50 into horizontal slices, or horizontal slice-formed sections, 60, 62, as shown in Fig. 9. The slice-formed sections 56, 58, 60, 62 may have a length m, and m may have a length between 1 pixel, and the maximum number of pixels in the given direction. It has however been found advantageously to apply either the horizontal or the vertical filtering, as the combined horizontal and vertical filtering at this stage seems to eliminate too many features. For both horizontal and vertical slice-formed sections 56, 58, 60, 62, the number of non-zero pixels in a slice-formed section 56, 58, 60, 62 should be below a predetermined first threshold to thereby eliminate intensity stemming from fractures or noise. Thus, the first filter 109 comprises determining the number of non-zero intensities in each slice-formed section 56, 58, 60, 62, i.e. pixels which have not been eliminated previously. If the number N of pixels with a non-zero intensity in a section 56, 58, 60, 62 is below the predetermined first threshold, all the pixels of that section 56, 58, 60, 62 are eliminated. Formally, the criterion may be written as:

wherein K_max is the first threshold and wherein the pixels of a section S are excluded, i.e. set to 0 when the criterion is fulfilled for the section.

The first threshold K_max may for the first filter 109 be determined based on the size of the slice-formed section S, which in the first filter 109 may be determined by the resolution of the image. The first filter 109 may be applied to the globally filtered image 14. and a further filtered image, such as a first further filtered image, is provided.

Applying the second filter 1 1 1 comprises segmenting the image into rectangular or square sections of sizes n x m pixels, as illustrated and described in relation to Fig. 5. In the second filter 1 1 1 , n and/or m may be between 150 and 450, such as between 250 and 350 such as 300.

The rationale behind the second filter 1 1 1 is the same as behind the first filter 109. If the number of non-zero pixels in a section 54 is below a

predetermined second threshold, then the identified pixels are likely to be scattered pixels which are not expected to represent fluid inclusions, and the intensity values of the identified non-zero pixels are set to zero.

Therefore, applying the second filter 1 1 1 comprises determining a number of non-zero intensities in each section 54. If the number N of non-zero intensities in a section 54 is below a predetermined second threshold, the second filter 1 1 1 excludes or eliminates all the pixels of that section 54.

Formally, if:

wherein K_max is the second threshold, then the pixels of that section S are excluded, i.e. set to 0.

Kmax for the second filter 1 1 1 may be determined based on the size of the section S. Hence,

K_max = f(n, m)

wherein f may be a linear function, such as a constant The second filter 1 1 1 may be applied to the first further filtered image resulting from the application of the first filter 109 to provide a further filtered image, such as a second further filtered image.

Applying the third filter 1 13 comprises segmenting the image into vertical slices 56, 58 with a width of n pixels, as illustrated and described in relation to Fig. 8. For the third filter 1 13, n may be between 10 and 100, such as between 25 and 75 such as 50.

A high sum of intensities along a quite narrow band, either vertically or horizontally may indicate a linear artifact, e.g. a fracture. Thus, a summed intensity within a vertical slice-formed section 56, 58 should be below a predetermined third threshold if the section should comprise a fluid inclusion and not e.g. a fracture. Therefore, the third filter 1 13 excludes, i.e. set to 0, all pixels within a vertical slice 56, 58 if:

∑ I 'p^Si.x > I ^lmax

\_max may be determined based on the size of the slice S, which for the third filter 1 13 is determined by the resolution of the image. The third filter further comprises subsequently to segmenting the image into vertical slice-formed section 56, 58 and applying the above criterion, segmenting the image into horizontal slice-formed sections 60, 62 with a width of n pixels, as illustrated and described in relation to Fig. 9, wherein n may be between 10 and 100, such as between 25 and 75 such as 50. It is envisaged that the third filter may also segment the image into firstly horizontal slice formed section and secondly into vertical slice formed sections.

The summed intensity within a horizontal slice-formed section 60, 62 should be below a predetermined threshold for the section to comprise fluid inclusions, since a high sum of intensities along a quite narrow band indicates a linear artifact, e.g. a fracture, rather than fluid inclusions. Thus, all pixels within a horizontal slice-formed section 60, 62 are eliminated by the third filter, i.e. set to 0, if:

wherein \_max is a third threshold intensity. \_max may for the third filter 1 13 be determined based on the size of the slice-formed section S, which may be determined by the resolution of the image.

The third filter 1 13 may be applied to the image, or the intensity map, resulting from the application of the second filter and provided a further filtered image, such as a third filtered image.

The fourth filter 1 15 comprises segmenting the image into rectangular or square sections 52 of sizes n x m pixels, as illustrated and described in relation to Fig. 5. In the fourth filter 1 15, n and/or m may be between 150 and 450, such as between 250 and 350 such as 300.

A summed intensity within a section 54 should be above a predetermined fourth threshold to provide an indication of the presence of fluid inclusions. Hence, if the predetermined fourth threshold is low, such that only areas with significant indications of fluid inclusions are considered to be fluid inclusions. Thus, by the fourth filter 1 15 all pixels within a section 54 are excluded, i.e. set to 0, if: ∑I *p^Si-x < I ^lmax

The predetermined fourth threshold, \_max may for the fourth filter 1 15 be determined based on the size of the section S.

The fourth filter 1 15 may be applied to the result of the third filter, i.e. to the third filtered image. In Fig. 2, the fourth filter 1 15 is applied to the result of the third filter 1 13.

An exemplary result of applying a detection method 100 according to Fig. 2 is seen in Fig. 10. It is seen on the resulting filtered image 18, that the only non- excluded areas are the fluid inclusions 10. Fig. 10a shows the filtered image as visualized by a heat map, including a colour code for the intensity values so that green colours correspond to high intensities, yellow colour to medium intensities and red colours to low intensities. Fig. 10b shows the heat map in black and white, thus dark green colours, corresponding to a high intensity within areas of lower intensity are seen as lighter spots, whereas a solid dark red colours, corresponding to a low intensity is seen as a solid dark colour.

In Fig. 1 1 image 26 shows a magnified region of image 6 as shown in Fig. 3, wherein the magnified region is the region identified by the detection method 100 having indications of fluid inclusions 28.

A scaling factor filter may be applied to the globally filtered image, or alternatively to any other filtered image. The scaling factor filter may comprise segmenting the image into rectangular or square sections, as illustrated and described in relation to Fig. 5. Fluid inclusions have, as described earlier, lower intensities than noisy regions and further lower densities. If the number N of non-zero intensities, i.e. pixels which have not already been excluded, in a section is greater than a predetermined threshold, all the pixels of that section are excluded by the scaling factor filter. Formally, if: N(¾_x > 0) > g^■ K max ' all pixels within the section S should be excluded, i.e. set to 0. g denotes a scaling factor to account for conditional handling of very large fluid inclusions and may take a value such as a value between 1 and 10, such as between 3 and 8, such as between 5 and 7, such as 6. gK_max is the scaling factor threshold and the scaling factor threshold may be determined based on the size of the section S.

In the above description, excluded pixels have been set to zero. However in other exemplary implementations, pixels may be excluded by setting the pixels to other values, such as a value not being a number or to a maximum pixel intensity, i.e. 255. In even other exemplary implementations, excluded pixels may be registered in a separate map or a vector.

Thresholds and other constants for each of the above described filters may be determined and fine-tuned based on empirical data. Fig. 12 shows a method 200 for validation of identified fluid inclusions. The validation method 200 comprises six validation criteria 202, 204, 206, 208, 210, 212, such as statistical criteria, for validating a resulting filtered image from a detection method 100. The validation criteria 202, 204, 206, 208, 210, 212 may be performed in a specific order, e.g. the order as illustrated, or they may be interchanged, or performed in parallel .

The validation method 200 comprises summed intensity criterion 202 defining that the resulting filtered image must have a summed intensity /_tot,res of at least a predetermined threshold /_min, formally:

/min may for a given resolution be between 130.000 and 250.000 such as between 170.000 and 210.000 such as 190.000. For example, /_min may be determined from the resolution of the resulting image. The validation method 200 comprises an intensity deviation criterion 204. The intensity deviation criterion 204 compares the standard deviation of a global intensity distribution of the received digital image <¾ , or initial image Oinit, to a predetermined threshold a_max. Formally: Ojnit *^■· 0~max-

The threshold a_max may be below 90 such as below 70 such as below 50 such as 40.

The validation method 200 furthermore comprises an intensity mean criterion 206. The mean criterion 206 compares the mean value of the intensity distribution of the resulting filtered image _res with the mean of intensity in the intensity distribution of the initial image μ,ηκ. The intensity mean criterion 206 is satisfied if _res is significantly less than μ_ιηΛ, with a p value less than a predetermined threshold a. a may be less than 0.001 such as below 0.0005 such as below 0.0001 . The validation method 200 comprises a mean-shift criterion 208. The mean shift criterion 208 compares the difference between the mean of intensity in the initial image μ,ηκ and in the resulting image _res with the standard deviation of intensity in the initial image. Formally: init - _res > |3CJ j_nit, wherein β is constant that may be between 1 and 10 such as between 1 and 5 such as 2.

The validation method 200 comprises a noise to signal criterion 210. The noise to signal criterion 210 compares the ratio of the standard deviation and mean of intensities in the initial image with a constant. In order to satisfy the noise to signal criterion 210, the ratio needs to be below a critical ratio C between 0 and 1 , such as between 0.2 and 0.6, such as 0.4. Formally:

^ < C.

^init The validation method 200 comprises a score criterion 212. The score criterion 212 is a combined score of different weighted criterion which is compared to a critical score Sc. Formally, in one embodiment the score may be defined as: -_Kl io_{g p} + κ₂ i) > 5_c,

wherein the weighting factors Ki , K₂, K₃, K4 and K₅ are real numbers. In one or more specific examples Ki to K₅ may be 1 , 3, 1 , -3 and 30, respectively.

Fig. 13 shows an automatic detection method 300 for detection of fluid inclusions. A sample is prepared for analysis 302. A digital image of the sample is obtained 304, e.g. by a light microscope. The image is analyzed using a detection method 306 according to the detection method 100 described in relations to Fig. 2. The result of the analysis 306 is validated using a validation method 308 according to the validation method 200 described in relations to Fig. 12. An output is received 310, which may be further inspected or provide for further analysis of the detected fluid inclusions.

Fig. 14 schematically illustrates an exemplary system 400 for automatic detection of fluid inclusions. The system 400 comprises a sample receiving unit 402, a microscope 406, a computer unit 410 and a post analysis unit 424. The computer unit 410 comprises a processing unit 412, a memory or storage 420 and a user interface 416.

A sample is placed in the sample receiving unit 402 and the microscope 406 obtains a colour image of the sample 404. The image is transmitted 408 from the microscope 406 to the processing unit 412 of the computer 410. The processing unit 412 performs a detection method in accordance with the detection method 100 described in relations to Fig. 2. Further, the processing unit 412 performs a validation of the result from the detection method in accordance with the validation method 200 described in relations to Fig. 12. At any time during the detection method or validation method, the processing unit 412 may read and/or write data 418 to the memory 420, either for storing results and/or for retrieving information, e.g. constants, settings etc.

The post analysis unit 424 receives the validated result 422 from the processing unit 412. The post analysis unit post examines 426 the identified fluid inclusions of the sample 402, e.g. by probing to determine the content of the inclusion.

The user interface 416 may be used to manage and control 414 the detection of fluid inclusions, e.g. by changing and/or setting constant values. The user interface 416 may also be used to choose whether or not to perform a next proposed step.

In one example, the image analysis is performed by receiving a bit map image in colour, and in step 1 , converting the recevied colour image to a greyscale image.

The conversion of the initial colour image to a greyscale image is done using a linear correlation between intensities of the red, green and blue channel in the colour image and the corresponding grey-scale intensity: Igrey = 0.2989 I red + 0.5866 Igreen + 0.1145 l_bhje

The three coefficients dictate the weights to the red, green and blue channels.

This algorithm may be implemented in C# in an extraction module of the source code.

The output of the conversion may be a .txt file of approx. 93MB for a picture containing approx. 5.000.000 pixels.

In step 2, blue-staining from epoxy glue and massive noisy regions may be removed (corresponding to application of the initial filter). Pixel intensities in regions with blue-staining and noise are typically between 75-175 (correspond to yellow colour) and importantly the density of pixels having this intensity is significantly higher compared to other regions of the image. In contrast fluid pixel intensities of fluid inclusions are lower (50-75) and surrounded by very high intensity pixels (>175).

The image is now segmented into quadratic sections S of each nxn pixels (e.g.: n=100). The sections containing blue-staining and noise are identified from the sum of pixel intensity l _ix. If

all pixel intensities in section S is set to 0. E.g.: /_max=1 ,700,000. Since fluid inclusions are surrounded by high-intensity pixels these will be kept during removal of blue-staining and noise.

In step 3, filtering based on the global intensity distribution is performed (corresponding to application of the global filter). Fluid inclusions predominantly contain pixels with intensities <100. Moreover from the intensity distribution of the received image, it may be seen that the fluid inclusions are in the lower tail hereof. Hence a very harsh filter is applied on the global image where all intensities !_pm greater than the p^f/7-quantile of the intensity distribution resulting from step 1 will be set to 0. E.g.: p=0.002 (i.e. the 0.2%-percentile).

In optional step 4 filtering is based on quadratic section-specific intensity distribution discrimination.

The image resulting from step 3 is now segmented into quadratic sections S of each nxn pixels (e.g.: n=75). If the number N of non-zero intensities is greater than a given threshold in section S all intensities in S are set to 0. Formally, if Ν(ΐξ_ίχ > 0) > g - K ^l _vmax all pixel intensities in section S are set to 0. E.g.: K_max=150 and g=6. The rationale for this step is that fluid inclusions are typically less intense than 'noisy spots', g is a scaling-factor related to conditional handling of very large inclusions. This conditional handling parameterized by the scaling factor g may be optional.

In step 5, filtering based on vertical section-specific intensity distribution discrimination is performed (corresponding to application of the first filter). The image resulting from step 4 is now segmented into vertical slices S each having a width of n pixels (e.g.: n=100). If the number N of non-zero intensities is less than a given threshold in section S all intensities in S are set to 0. Formally, if

all pixel intensities in section S are set to 0. E.g.: K_max=200.

The rationale for this step is to remove vertical features such as fractures. This may be done horizontally as well, however it has been found that the application of two successive filters may lead to too much intensity elimination, thus performing either horizontal or vertical filtering may be beneficial.

In step 6, filtering is based on quadratic section-specific intensity distribution discrimination (corresponding to application of the second filter).

The image resulting from step 5 is now segmented into quadratic sections S of each nxn pixels (e.g.: n=300). If the number N of non-zero intensities is less than a given threshold in section S all intensities in S are set to 0. Formally, if N{l_p ^s _ix > 0) < Κ_Ί max all pixel intensities in section S are set to 0. E.g.: K_/T?ax=200.

In step 7, removal of vertical and horizontal fractures is performed (corresponding to application of the third filter).

In step 7(A), the image resulting from step 6 is now segmented into vertical slices S each having a width of n pixels (e.g.: n=50). If

all pixel intensities in section S are set to 0. E.g.: I_max= 280,000.

Step 7(B), the image resulting from step 7A is now segmented into horizontal slices S each having a width of n pixels (e.g.: n=50). If

all pixel intensities in section S are set to 0. E.g.: I_max= 280,000.

In step 8, post-processing removal of fragments is performed (corresponding to the application of the fourth filter)

The image is now segmented into quadratic sections S of each nxn pixels (e.g.: n=300). The sections containing blue-staining and noise are identified from the sum of pixel intensity l _ix. If

pix max all pixel intensities in section S are set to 0. E.g.: /_max=600.

In step 9, potential fluid inclusions are identified.

After the final processing step (step 8) al spots left in the resulting image are potential fluid inclusions. The software will zoom in on the region containing the potential fluid inclusions. The region may be enlarged by n pixels in each direction around the pixels closest to the edge of the image in each direction to include potential fluid inclusions filtered away during the processing steps. Thus, if fluid inclusions are lost during the processing, these can still be visualized by its neighboring identified fluid inclusions. The candidates may not be accepted as being fluid inclusions at this point since a number of criteria for the resulting picture must be fulfilled (explained below, step 10).

Step 10: Validation of identified potential fluid inclusions

In order for the identified potential fluid inclusions to be accepted as being fluid inclusions one or more of the following six statistical criteria must be fulfilled, and typically, it is advantageously to accept the potential fluid inclusions as fluid inclusions if all criteria are fulfilled:

1 . The resulting image must have a summed intensity /_ioi._res of at least l_min (e.g.: =190,000)

2. The standard deviation of a global intensity distribution of the received digital image <¾ must be below a rounded (for example to 0 decimals) predetermined threshold a_max.i_nit (e.g.:

3. the mean value _res of the intensity distribution in the resulting filtered image after processing (i.e. after step 8) must be significantly lower than the mean value of the intensity distribution of the initial image μ,ηκ, i.e. of the initial intensity map (after step 1 ), at a significance level of a (e.g.: α=0.0001 ) using the one-sided (test for down-shift of mean) Welch T-test. The p-value must therefore be less than a, p <

The mean-shift from the intensity distribution of the initial intensity map to the intensity distribution of the resulting intensity map, i.e. the difference between the mean value of the intensity in the initial image Minit and in the resulting image μ_Γβ8, must be greater than β (e.g.: β=2) times the standard deviation of the initial intensity distribution:

The noise-to-signal ratio of the initial image, i.e. the ratio of the standard deviation and mean of intensities in the initial image, must be below a critical ratio C (e.g.: C=0.4). C must be between 0 and 1 . init < c

A score defined as below must be greater than a critical score S_c (e.g.

- logG + 3 > S_c

• -log(p) is a measure of the magnitude of the mean shift is a measure of how much smaller the

noise-to-signal ratio is compared to the pre-set critical ratio

• (?ML \ j_{s a} measure of the initial signal-to-noise ratio, i.e. how good is the signal-to-noise ratio

• (?ML \ j_{s a} measure of the noise-to-signal ratio, i.e. how bad is the signal to noise ratio

• ^res - 1 is a measure of how different the initial and final post-processed image is, i.e. are enough intensity removed after the processing (which is expected to be the case)

The parts of the score formula are weighted by 1 :3:1 :(-3):30. Note that a bad signal-to-noise ratio is penalized 3 times more than a good signal-to-noise ratio is amplified which is not amplified since the

'amplification factor' is 1 .

Preferably, if all six criteria are met identification of fluid inclusions is validated and the inclusions are accepted.

Although particular embodiments of the present inventions have been shown and described, it will be understood that it is not intended to limit the claimed inventions to the preferred embodiments, and it will be obvious to those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the claimed inventions. The specification and drawings are, accordingly, to be regarded in an illustrative rather than restrictive sense. The claimed inventions are intended to cover alternatives, modifications, and equivalents. LIST OF REFERENCES

2 received image

4 region of fluid inclusions

6 converted image

8 noisy regions

10 fluid inclusions

12 initially filtered received image

14 globally filtered image

16 scattered pixels

18 resulting filtered image

30 image intensity distribution

32 pixel intensity

34 count of pixels

36 low intensity region

38 medium intensity region

40 high intensity region

50 schematic exemplary image

52 segmenting an exemplary image

54 a section of an exemplary image 56 vertical slicing of an exemplary image

58 a vertical slice of an exemplary image

60 horizontal slicing of an exemplary image

62 a horizontal slice of an exemplary image

100 detection method 102 intensity conversion

104 initial filter

106, 108 global filter

109 first filter

1 10 set of filters

1 1 1 second filter

1 13 third filter

1 15 fourth filter

200 validation method

202 summed intensity criterion

204 intensity deviation criterion

206 intensity mean criterion

208 mean-shift criterion

210 noise to signal criterion

212 score criterion

300 automatic detection method

302 sample preperation

304 image obtainment

306 detection method

308 validation method

310 output of results

400 automatic detection system

402 sample receiving unit

404 otaining a colour image of the sample 404 406 microscope

408 transmittal of colour image

410 computer

412 processing unit

414 user interface control

416 user interface

418 dataflow to/from memory

420 memory/storage

422 validated result

424 post analysis unit

426 post examination of sample

Claims

1 . A method of automatic detection of fluid inclusions in crystalline materials, the method comprising

receiving at least one digital image of a crystalline material

-determining global image intensity properties of each pixel of the received digital image

- applying one or more global image filters on the determined global image intensity properties to provide a first filtered image,

-successively applying a set of filters by

segmenting the first or a further filtered image into a plurality of sections,

applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections,

providing a further filtered image, providing a resulting filtered image, and

based on the resulting filtered image, identifying fluid inclusions in the at least one received image.

2. A method according to claim 1 , wherein the received digital image comprises a plurality of pixels, each pixel having one or more intensity values.

3. A method according to any of the previous claims, wherein the method further comprises the step of segmenting the received digital image into a plurality of initial sections, determining a sum of intensity values for each of the plurality of initial sections, applying an initial filter on the determined sum of intensity values for each of the plurality of initial sections, and providing an initially filtered received image.

4. A method according to claim 3, wherein the initial filter comprises determining a sum of pixel intensities within each of the initial sections, and for each initial section wherein the sum of pixel intensities is less than an initial threshold intensity value, setting all pixel intensities in that initial section to zero to thereby filter out blue staining and/or noise.

5. A method according to any of the previous claims, wherein for each filter, the segmenting of the filtered image and/or of the received image is predetermined.

6. A method according to claim 5, wherein a size of the plurality of sections is predetermined.

7. A method according to any of the previous claims, wherein the global image filter comprises

calculating a global intensity distribution of the received image

setting a global pixel threshold intensity based on the global intensity distribution of the received image, and

for all intensities of the received image being greater than the global pixel threshold intensity, setting the intensity to zero.

8. A method according to any of the previous claims, wherein the set of filters comprises a filter in which if a number of pixels having non-zero intensities within a section is less than a threshold, then all intensities in that section is set to zero.

9. A method according to claim 8, wherein a first filter is implemented with a first threshold and wherein the filtered image is segmented into vertical and/or horizontal slice-formed sections, each slice-formed section having a predetermined first size.

10. A method according to claim 8, wherein a second filter is implemented with a second threshold and wherein the filtered image is segmented into rectangular sections of a predetermined second size.

1 1 . A method according to any of the previous claims, wherein the set of filters further comprises a third filter, wherein the filtered image is segmented into vertical and/or horizontal slice-formed sections of a predetermined third size, and wherein for each slice-formed section in which a sum of pixel intensities is larger than a third threshold, then the pixel intensity in that section is set to zero.

12. A method according to claim 1 1 , wherein the third filter is sequentially applied to vertical slice-formed sections and horizontal slice-formed sections.

13. A method according to any of the previous claims, wherein the set of filters further comprises a fourth filter and wherein the filtered image is segmented into rectangular sections, and wherein for each rectangular section in which a sum of pixel intensities is less than a fourth threshold, then the pixel intensity in that section is set to zero.

14. A method according to any of the previous claims, wherein the method further comprises providing positions of the identified fluid inclusions in the at least one digital image.

15. A method according to any of the previous claims, wherein identified fluid inclusions are validated if one or more of the following criteria are fulfilled: a) the resulting filtered image has a summed intensity /_ioi._res of at least l_min;

I tot. res— Imin b) the standard deviation of a global intensity distribution of the received digital image <¾ is below a threshold intensity distribution a_max.init_;

c) the mean value of a global intensity distribution in the resulting filtered image is less than the mean value μ_ηΛ of the global intensity distribution in the received digital image at a significance level of a, so that a p-value, p, is less than a;

p< a d) the mean-shift from the mean value μ,ηκ θί the global intensity distribution in the received digital image to the mean value of the intensity distribution of the resulting filtered image jL futer is greater than β times the standard deviation of the global intensity distribution of the received image -init ~ ^filter β^σίηίΐ e) The noise-to-signal ratio of the receive digital image must be below a critical ratio C (e.g.: C=0.4), defined as the ratio between the standard deviation of the global intensity distribution of the received digital image init and the mean value μ_ηΛ of of the global intensity distribution in the received digital image where C is between 0 and 1 ,

f) a score S is greater than a critical score S_c, wherein S is defined as:

^ ^init

wherein Ki , K2, K₃, K and K₅ are integer numbers,

• -log(p) is a measure of the magnitude of the mean- shift a measure of how much smaller the

noise-to-signal ratio is compared to the pre-set critical ratio

j_{s a measure} of the initial signal-to-noise

ratio

noise-to-signal ratio,

how different the initial and min

final post-processed image is,

16. A method according to any of the previous claims, wherein the inclusions are between 0.1 and 20μηη.

17. A method according to any of the previous claims, wherein the method further comprises analysing an identified or validated fluid inclusion.

18. A system for automatic detection of fluid inclusions in crystalline materials, the system comprising

a processor configured to receive at least one digital image of a crystalline material, to determine global image intensity properties of the received digital image and applying one or more global image filtering criteria on the determined global image intensity properties to provide a first filtered image, and a storage for storing the at least one digital image and at least temporarily the first and further filtered images,

the processor being further configured to successively apply a set of filters by segmenting the first or a further filtered image into a plurality of sections, applying a filter from the set of filters on one or more determined intensity properties for each of the plurality of sections, and providing a further filtered image,

the processor being configured to provide a resulting filtered image, and based on the resulting filtered image, identify fluid inclusions in the at least one received image.

19. A system according to claim 18, wherein the processor is configured to provide positions for the identified fluid inclusions.

20. A system according to claim 19, the system further comprises an interface for providing the fluid inclusion positions to a camera, microscope or an analyser for facilitating further investigation of the identified fluid

inclusions.

21 . A computer program comprising program code means for performing the steps of any one of the claims 1 to 17 when said computer program is run on a computer.

22. A computer readable medium having stored thereon program code means for performing the method of any one of the claims 1 to 17 when said program code means is run on a computer.