SHERPA: an image segmentation and outline feature extraction tool for diatoms and other objects
© Kloster et al.; licensee BioMed Central Ltd. 2014
Received: 15 May 2014
Accepted: 27 May 2014
Published: 25 June 2014
Light microscopic analysis of diatom frustules is widely used both in basic and applied research, notably taxonomy, morphometrics, water quality monitoring and paleo-environmental studies. In these applications, usually large numbers of frustules need to be identified and/or measured. Although there is a need for automation in these applications, and image processing and analysis methods supporting these tasks have previously been developed, they did not become widespread in diatom analysis. While methodological reports for a wide variety of methods for image segmentation, diatom identification and feature extraction are available, no single implementation combining a subset of these into a readily applicable workflow accessible to diatomists exists.
The newly developed tool SHERPA offers a versatile image processing workflow focused on the identification and measurement of object outlines, handling all steps from image segmentation over object identification to feature extraction, and providing interactive functions for reviewing and revising results. Special attention was given to ease of use, applicability to a broad range of data and problems, and supporting high throughput analyses with minimal manual intervention.
Tested with several diatom datasets from different sources and of various compositions, SHERPA proved its ability to successfully analyze large amounts of diatom micrographs depicting a broad range of species. SHERPA is unique in combining the following features: application of multiple segmentation methods and selection of the one giving the best result for each individual object; identification of shapes of interest based on outline matching against a template library; quality scoring and ranking of resulting outlines supporting quick quality checking; extraction of a wide range of outline shape descriptors widely used in diatom studies and elsewhere; minimizing the need for, but enabling manual quality control and corrections. Although primarily developed for analyzing images of diatom valves originating from automated microscopy, SHERPA can also be useful for other object detection, segmentation and outline-based identification problems.
KeywordsDiatom Segmentation Outline Elliptic Fourier analysis Shape descriptors Morphometrics Automated slide scanning
Diatoms are a group of photosynthetic protists producing uniquely ornamented and diversely shaped silicate shells . They are present in all aquatic and wet habitats and, with an estimated 105 species, they represent the most species rich algal group . Diatom assemblage composition reflects the abiotic and biotic features of their respective habitats, and is widely used for making inferences about environmental conditions in water quality monitoring and paleontology . Due to a combination of traditional and practical reasons, the most widely applied method for diatom investigations is based on light microscopic analysis of so called permanent slides, prepared using the silicate frustules after cleaning them of organic material .
Size and shape distributions of diatom populations are measured and analyzed in a number of different fields, including taxonomy [4–8], ecology [9–12], and paleontology [13–16]. In such studies, dozens to hundreds of specimens are routinely investigated from each of several slides, and measurements are usually performed by one of the following methods: 1) through an ocular micrometer directly on images seen in the microscope by the investigator ; 2) as manual (mostly, length) measurements on digital live images presented on a computer screen [4, 16]; 3) as manual (again mostly, length) measurements on saved digital images using general purpose image analysis software ; 4) combination of manual measurements and measurements obtained by custom-developed macros or extensions of general purpose image analysis software like ImageJ  or Optimas [5, 7].
There is a considerable methodological gap between these approaches and the sometimes rather sophisticated methods which have been applied to diatoms in the image analysis literature for instance in the project ADIAC , or by others including [19–21]. Much of the experience gained in diatom image analysis studies should in principle be transferable to diatom morphometrics and would have the potential to speed up the latter and make it more accurate and reproducible. However, these methods have remained practically inaccessible to diatomists due to a lack of publicly available and user friendly implementations of image processing and analysis methods suitable for diatom analyses. Most of the diatom image analysis literature does not explicitly state which software tool or framework was used for implementing the applied methodology. Although this practice reflects a focus upon algorithms and methods, as opposed to software, and is probably well suited for readers with their main area of expertise lying in computer science and image analysis, translating these methodological experiences into routinely practicable workflows has remained a challenge beyond the qualification of most, if not all, diatomists, as illustrated by the almost complete lack of reports on re-use of these methods beyond the groups which developed them. The only case known to us where implementations of individual algorithms have been made available publicly is represented by the small collection of MATLAB and C source code files available under . However, even these only represent fragments of a practically applicable analysis workflow and are virtually inaccessible to most diatomists (at least to the overwhelming subset lacking familiarity with MATLAB/C programming).
Several of the individual algorithms tested and applied in diatom image analyses in the above cited works represent standard image analysis methods, with widely available implementations in general purpose image analysis software like ImageJ . Thus, it could be argued that such software should also be perfectly suited for the needs of diatomists. However, in our experience, whereas for instance ImageJ can be useful for processing and analyzing individual diatom images or small collections thereof, building a workflow for high throughput work with it requires serious programming capabilities, a reason probably hindering the use of such software in diatom studies. For instance, a number of segmentation algorithms can successfully be applied to diatom valves, but it is often found that a different method works best for different objects, depending not only on valve structure (and thus, also taxonomy) but also upon minor details of how the object lies relative to the focal plane and to neighboring objects . Whereas one can easily apply a handful different segmentation algorithms to an image in for instance ImageJ, deciding which one gives best results in a case-by-case manner can be challenging. Doing so programmatically to enable batch processing of large numbers of images with minimal manual interaction would go beyond the capabilities of most non-image-analysis-expert users of ImageJ. Since diatom images are notoriously difficult to segment due to the optical properties of the silicate shells (low contrast, strong halo around outline, huge structural and shape diversity), chaining together individual analysis steps to an automated workflow also requires some kind of quality control. Differentiating objects of interest (diatom frustules, or, in particular cases, frustules of a particular group of diatoms) from other objects found by segmentation methods (sediment particles, debris, non-target species) would also require considerable programming skills to implement in ImageJ.
The outline represents a rather information rich aspect of the morphological variability of diatom frustules, and its shape and size contains substantial taxonomic and life cycle related information especially in the case of pennate diatoms (even if it has to be noted that diatom identification at the species level is mostly impossible based on outline shape alone). The main approaches for quantitative characterization of outline shapes in diatom morphometrics have included the use of simple heuristic shape descriptors like rectangularity , ellipticity, compactness [18, 24]; Legendre-polynomials ( and the large body of literature cited therein); Fourier descriptors [18, 25, 26]; and landmarks and semi-landmarks [8, 27–31]. Although further methods have been developed, some specifically for diatoms, notably the segment shape analysis approach  successfully applied in , these have not become widely used. General purpose morphometrics software [33, 34] is available for landmark and semi-landmark digitization and analysis, but using such software, landmark points need to be digitized individually and manually, hindering high throughput analyses. For other types of outline descriptors, some software support is available (see e.g. examples for software tools capable of calculating elliptic Fourier coefficients under ), but again not as part of routinely applicable workflows supporting the analysis of large numbers of images.
With SHERPA presented in the present paper, we address these gaps and introduce an easy-to-use tool for segmenting and analyzing light microscopic images of diatom frustules, and for extracting a number of outline features useful for diatom morphometrics (but potentially in other fields as well). Our goals were to develop a tool that implements 1) a full image analysis workflow from image segmentation to outline feature extraction, specifically adapted to diatom images, but potentially useful for other objects where outline shape is informative; 2) multiple segmentation methods and an automated selection of the best result for each segmented object; 3) matching of object outlines against a set of template outlines to enable both taxonomically selective as well as broader analyses; 4) object scoring and ranking to support quality checking; 5) extraction of a wide range of outline shape descriptors for further analyses; 6) supporting processing of large batches of images by minimizing the need for manual interaction, but leaving the possibility for it in case it should be required, e.g. to correct outlines for diatom valves with minor overlaps with neighboring objects. Software implementing statistical and/or machine learning methods for exploration, analysis, and classification of large multivariate data sets is widely available both commercially and free of charge for users at a wide range of levels of computer fluency (ranging for instance, from the easy-to-use PAST  or JMP  to the more challenging, but also more versatile statistical analyses systems like R  or SPSS ). Accordingly, we decided to not include this functionality in our tool but rather generate output that can be loaded for downstream analyses into the user’s statistical tool of choice.
This way, extensive image collections can be processed in a fully automated manner or with minimal manual intervention. Irrelevant data, originating from debris, damaged or unwanted objects, can be sorted out with little or no user intervention at all, while relevant objects are identified and measured. The exported morphometric descriptors allow for a detailed and specific analysis based on tools like R , and questions about variation in outline shape and size can easily be investigated.
Comparison of features of SHERPA and ImageJ
Integrated workflow for segmentation, identification and measurement of objects
Automatic combination of multiple segmentation methods
Automatic combination of multiple contour optimization methods
Convexity defect measures
Ranking of segmentation results
Quick interactive review of results
In order to create a low level entry point for novice users, extensive documentation is provided along with the software, including a comprehensive manual, a quick-start guide, a tutorial on how to achieve suitable settings in a straightforward way, and a technical description of the analysis process and extracted morphometric features.
SHERPA was developed for Windows7 64 Bit using C#/.NET 4.0. Most image processing functions are realized based on OpenCV 2.4.2 , whose DLLs are wrapped for .NET by Emgu CV 2.4.2 , and on ITK 4.2  called via external executables. "Microsoft .NET Framework 4"  and the "Microsoft Visual C++ 2010 SP1 Redistributable Package (×64)"  have to be installed prior to running SHERPA. A 32 Bit version of SHERPA is available, but its usage is not recommended because it might run out of memory resources when analyzing large amounts of data.
Image data to be analyzed can depict objects either as dark structures on bright background (like obtained e.g. using bright field microscopy) or as bright structures on dark background (like obtained e.g. using dark field microscopy). Objects are identified by shape information. For proper results, object outlines should be focused as precisely as possible. Minor blurring will affect the accuracy of outline detection, while extensive fuzziness might impede usable results. For an optimal identification yield the sample density should be sparse without overlapping objects.
Image data is converted into shape information by applying a consecutive set of image processing functions:
Noise reduction can be performed by applying Gaussian or median filtering.
Shape detection is accomplished by following each object outline using an algorithm by Sklansky . The outer object contour is the starting point for subsequent analysis steps.
Shape processing and analysis
Shapes derived from image processing might be flawed due to segmentation problems or overlapping objects, and they can depict anything from objects of interest to debris and foreign particles. To increase the yield of usable results and to sort out irrelevant data, shapes can be optimized and are evaluated according to their chance of depicting a relevant object.
Manual rework is an option if a shape is distorted due to segmentation flaws, but the corresponding object is essential as a valid result. SHERPA offers functions for redrawing a contour like in a painting program, for smoothing it and for applying morphological operators (see above) with individual settings to it, as well as to expand the outline to its convex hull.
Rating and ranking
Matching and quality indicators used for ranking
EFDIs Match with Template
Hu Match for EFDIs Template
Matching between the Hu invariants  of the object and the template which matches best according to EFDIs.
Morphological Operator used to improve the object contour. If an optimization was applied to derive a shape, its ranking is degraded, because the resulting outline might be inaccurate.
Standard Deviation of inner 50%
Standard deviation of the gray level distribution within the object boundaries. Only the inner 50% of the area are analyzed. This way, diatom valves, normally containing striae/costae/areolae, can be distinguished from empty girdle bands which can produce good outline matching but have a homogenous interior.
Ratio between object width and height. Usually objects of a certain type have a ratio within a certain range.
Estimation of the object contour smoothness. The actual object outline usually is quite smooth, especially for diatom valves, whilst contours distorted by segmentation inaccuracies or failures usually are rough. The ratio between the outline perimeter and that of the outline smoothed by a Gaussian filter provides information about the contour smoothness.
Heuristic descriptor "formfactor" 
Convexity defect measures used for ranking
"Convexity Defection Factor", depicts the percentaged difference between area resp. perimeter of contour and convex hull 
The "Percent Concave Area Fraction" compares the areas of contour and convex hull .
For the "Convex Hull Maximum Distance Factor" each convexity defect’s maximum distance between contour and convex hull is calculated. For distances larger than pixelwidth the squares of the distances are summed up to the CHMDF .
Ratio of CDF of object and template
Ratio of PCAF of object and template
Ratio of heuristic descriptor "compactness" between object and template shape
Matching and quality indicators rate the matching between shape and template and some properties which help to distinguish objects of interest from irrelevant ones, like e.g. width/height-ratio and standard deviation of the texture gray levels within the central part of the object (see Table 2).
Convexity defect measures (CDMs) are calculated based on differences of area and/or perimeter between a contour and its convex hull, the latter being the smallest area which encloses the contour without containing any concave parts.
This approach will not work for objects which naturally contain concave parts. If the data contains convex as well as concave objects, SHERPA’s feature "Use Convexity" can be activated. In this case, only if the best matching template is convex, CDMs are evaluated by their absolute values derived from the respective object shape (like when using "Force Convexity"). If the best matching template is concave, some CDMs plus the heuristic descriptor "compactness"  of the object will be compared to those derived from the best matching template (see Table 3, "Relative measures").
When the set of objects to be detected contains both convex and concave outlines and convexity analysis is employed (i.e. "Use Convexity" or "Force Convexity" is enabled), the template set should be composed with special care. The situation to be avoided is that the best match of a concave object becomes a convex template, which can happen if no proper concave template is provided. In this case, the object convexity will be judged by absolute values even though it is concave, which will result in a failure of convexity defect measures and hence in a poor ranking.
If neither "Use Convexity" nor "Force Convexity" are activated, only a relative comparison of some CDMs between object and template plus an evaluation of the form factor takes place, regardless if the best matching template is convex or concave. The object’s CDMs are not judged directly. This is usually a good choice if it is not known in advance if all relevant objects are convex and/or there is no extensive library of templates yet.
It should be noted that detection of segmentation flaws is much less accurate when an object’s convexity defect measures are compared to those of the template instead of being judged by their absolute values. So if only convex objects are of interest, choosing "Force Convexity" will provide a more precise ranking and might save some manual reviewing.
Review, rework and selection of results
Name of feature
Path to raw image data file
Object width (along major axis)
Object height (perpendicular to major axis)
Rotation angle of the major axis
Segmentation method used to derive the object shape
Optimization method applied to the object shape
Best Template (EFDIs)
Path to the best matching template (according to matching of elliptic Fourier descriptor invariants)
Template Difference (EFDIs)
Value for matching of elliptic Fourier descriptor invariants between object and best matching template
Hu-match for best EFDIs-Template
Value of matching of Hu invariants between object shape and best matching template
Standard deviation of texture gray levels within the inner 50% of the object boundaries
Aspect ratio of the object shape
Smoothed Perimeter Ratio
Ratio between the perimeters of the smoothed and the original contour; smoothing is performed by Gaussian filtering of the contour coordinates.
Number of fulfilled quality indicators
Template is convex
Indicator showing if the best matching template is convex
Convexity is used
Indicator showing if convexity was judged directly to calculate convexity indicators (use of absolute convexity measures)
Convexity by perimeter
Convexity by area
Convexity defect measure
Convexity defect measure
Convexity defect measure
Ratio of CDF between object and template
Ratio of PCAF between object and template
Ratio of heuristic descriptor "formfactor" between object and template
Convexity Defect Index
Number of fulfilled absolute or relative convexity indicators
Ranking for object shape, i.e. estimation of quality and relevance of result
Name of the file containing the image data cropped to the object area
Contour Image top left Corner
Coordinates of the top left corner of the cropped object image with respect to the raw data
Image Moments (mu)
Image moments of the object shape
Hu Invariants (Hu)
Hu-Invariants of the object shape
Results and discussion
For the following analyses, bright field micrographs of valves of different diatom species and from different sources were analyzed. All results were produced without manually reworking or resorting detected shapes, relying solely on SHERPA’s automated functions for segmentation, contour optimization and result ranking.
To facilitate use of SHERPA for generic diatom recognition and analysis, we prepared a library covering a wide range of diatom outline shapes, containing about 450 templates. This compilation is mainly based on the outline shape classification scheme and accompanying diagrams from Barber & Haworth , Fragilariopsis data sets from a surface sediment sample , and upon the extensive ADIAC diatom image database available online , although the ADIAC data is not fully covered by the current template library. For the latter two, SHERPA was used for image segmentation to detect shapes previously not represented in the template set: Shapes with a poor template matching value were screened manually. If they were depicting relevant valves and segmentation quality was satisfactory, they were converted into additional templates employing the built-in functions of SHERPA. Because diatom shapes vary widely among taxa, as well as during the life cycle of even a single taxon, it is crucial to check the presence of a representative set of templates for taxa of interest when using SHERPA for analyzing a particular type of diatom samples.
Sellaphoradata as example for identification accuracy
To demonstrate the usability of SHERPA, we analyzed a set of images from one of the classical model taxa of diatom microdiversity, the Sellaphora pupula (Kützing) Mereschkowsky complex s.l. S. pupula has been known as a morphologically highly variable diatom species during most of the 20th century. However, Mann and colleagues demonstrated in a series of papers (cumulating in ) that sympatric demes of this diatom "species" formed reproductively isolated groups, that could also be diagnosed using molecular markers and also differed in minute morphological/morphometric features, including (but not limited to) minor differences in their valve outlines. In their 2004 investigation , Mann et al. used Legendre-polynomials and contour segment analysis for comparing outline morphology of six S. pupula demes (since that study, also formally recognized as distinct species). They made the images upon which the analyses were based publicly available , which we used in this analysis.
Results analyzing 383 images  depicting Sellaphora valves (plus one centric diatom)
Identified as Sellaphora pupula
Identified as other1)
Results having a ranking above 2 are not listed, because they were caused by partly unfocused outlines, overlapping objects or debris and would have needed manual inspection and reworking.
Fragilariopsisdata as example for segmentation quality
As a typical data set, 773 micrographs originating from sediment core PS1768-8  and mainly showing Fragilariopsis kerguelensis, plus broken valves, debris and overlapping objects, were analyzed. The data was obtained using a Metafer slide scanning system (Metasystems, Altlussheim, Germany), applying the implemented autofocus and stacking functions. Because not all valves were lying parallel to the focal plane, outlines were partly out of focus or blurred despite of stacking. Since the outline of F. kerguelensis is completely convex, SHERPA’s "Force Convexity" feature was used to improve judging of segmentation quality.
Again, the full template set covering a broad range of diatom species was used. Although Fragilariopsis valves were mostly identified correctly, some were assigned to templates of other similarly shaped species, and some correctly identified valves of other species were present. Undamaged valves could successfully be distinguished from artifacts like broken ones or debris. In some cases, objects like girdle bands or spherical structures were identified as relevant valves (usually at a ranking index 2 or worse), because of their shape similar to those of other diatom species in the template library. This problem can be overcome by using only Fragilariopsis templates.
Results for Fragilariopsis data for different combinations of segmentations methods and contour optimization
RATS (σ = 3)
RATS (σ = 1-11)
Canny edge detector
Ranking 0 total
Ranking 1 total
Ranking 2 total2)
Total ranking 0 to 2
Comparison of segmentation methods
With MAX = maximum, MIN = minimum value for a feature (area, perimeter, etc.) when using multiple segmentation methods.
Further analysis using R
Besides improving performance, the next steps in SHERPA’s development will concern the analysis of texture und structural features to improve versatility and identification specificity.
SHERPA provides a useful tool for diatom identification and morphometrics, enabling mass screenings, since it greatly reduces the amount of work needed to be performed by human interaction. Manual revision required for best results can be accomplished in a quick and effective manner, supported by a ranking based on matching and quality indicators.
The degree of identification reliability reflects both the range of templates used and the diversity present in the analyzed samples. In spite of depending solely on outline shape, good identification accuracy can be reached using customized template sets. Combining multiple segmentation methods improves the identification rate without significantly impairing result accuracy, and, combined with contour optimization, even objects showing segmentation artifacts can be analyzed successfully. For convex shapes, convexity defect measures provide an effective way to judge segmentation quality, hence allowing identification of flawed object outlines.
The approach of restricting SHERPA to the identification of relevant objects and the calculation of their morphometric features enables an adaptation to specific problems/target taxa. Downstream analyzes or classification can be performed using widely available commercial or free statistical software tools, e.g. "R".
Availability and requirements
Project name: SHERPA.
Project home page: http://www.awi.de/sherpa.
Operating system(s): Windows7 64 Bit (32 Bit version available).
Programming language: C#.
Other requirements: .NET 4.0.
License: Freeware, royalty-free, non-exclusive.
Any restrictions to use by non-academics: none.
MK started developing SHERPA as part of his master thesis (at that time called "DiatoMorphoTo") at the HSEL, supervised by GK and in collaboration with BB. Since graduation he works at the Friedrich Hustedt Diatom Study Centre, AWI, under supervision of BB to develop SHERPA. He mainly works at the interface between biology and informatics, focusing on image processing, data visualization and automation.
GK is a professor in bioinformatics and has been working for more than 15 years in the areas of genome analysis, microscopy, image processing, image interpretation and development of bioinformatics methods for genome and proteome analysis. His recent works regard high performance reconstruction of structures from high resolution image stacks of extensive microscopic objects, and limited three-dimensional reconstruction from stereoscopic images of biological tissues and organisms. GK is also the author of E.L.M.I. (Expert System for Light Microscopy).
BB is a diatomist / bioinformaticist, curator of the Hustedt Diatom Study Centre. His research currently focuses on taxonomy, biogeography and morphometrics of Antarctic diatoms.
Tool for "Shape recognition, processing and analysis"
Convexity defect measures
Elliptic Fourier descriptors
Elliptic Fourier descriptor invariants
"Convexity defection factor", a convexity defect measure
"Percent concave area fraction", a convexity defect measure
"Convex hull maximum distance factor", a convexity defect measure.
Thanks to Rainer Gersonde for providing the slides from sediment core PS1768-8 used for the Fragilariopsis test, and to Nike Fuchs and Fabian Altvater for scanning and sorting the Fragilariopsis images used.
- Round FECRM, Mann DG: The diatoms. Biology and morphology of the genera. 1990, Cambridge, UK: Cambridge University PressGoogle Scholar
- Mann DG, Vanormelingen P: An inordinate fondness? The number, distributions, and origins of diatom species. J Eukaryot Microbiol. 2013, 60 (4): 414-420. 10.1111/jeu.12047.View ArticlePubMedGoogle Scholar
- Smol JP, Stoermer EF: The diatoms: applications for the environmental and earth sciences. 2010, Cambridge, UK: Cambridge University PressView ArticleGoogle Scholar
- Abarca NJR, Zimmermann J, Enke N: Does the cosmopolitan diatom Gomphonema parvulum (Kützing) Kützing have a biogeography?. Plos One. 2014, 9 (1): e86885-10.1371/journal.pone.0086885.View ArticlePubMed CentralPubMedGoogle Scholar
- Droop SJM, Mann DG, Lokhorst GM: Spatial and temporal stability of demes in Diploneis smithii/D-fusca (Bacillariophyta) supports a narrow species concept. Phycologia. 2000, 39 (6): 527-546. 10.2216/i0031-8884-39-6-527.1.View ArticleGoogle Scholar
- Kingston JC, Pappas JL: Quantitative shape analysis as a diagnostic and prescriptive tool in determining Fragilariforma (Bacillariophyta) taxon status. Nova Hedwigia Beih. 2009, 135: 103-119.Google Scholar
- Mann DG, McDonald SM, Bayer MM, Droop SJM, Chepurnov VA, Loke RE, Ciobanu A, du Buf JMH: The Sellaphora pupula species complex (Bacillariophyceae): morphometric analysis, ultrastructure and mating data provide evidence for five new species. Phycologia. 2004, 43 (4): 459-482. 10.2216/i0031-8884-43-4-459.1.View ArticleGoogle Scholar
- Poulickova A, Vesela J, Neustupa J, Skaloud P: Pseudocryptic diversity versus cosmopolitanism in diatoms: a case study on Navicula cryptocephala Kütz. (Bacillariophyceae) and morphologically similar taxa. Protist. 2010, 161 (3): 353-369. 10.1016/j.protis.2009.12.003.View ArticlePubMedGoogle Scholar
- Crawford RM, Hinz F, Rynearson T: Spatial and temporal distribution of assemblages of the diatom Corethron criophilum in the Polar Frontal region of the South Atlantic. Deep-Sea Res Pt II. 1997, 44 (1–2): 479-496.View ArticleGoogle Scholar
- Jewson DH, Granin NG, Zhdanov AA, Gorbunova LA, Bondarenko NA, Gnatovsky RY: Resting stages and ecology of the planktonic diatom Aulacoseira skvortzowii in Lake Baikal. Limnol Oceanogr. 2008, 53 (3): 1125-1136. 10.4319/lo.2008.53.3.1125.View ArticleGoogle Scholar
- Jewson DH, Granin NG, Zhdarnov AA, Gorbunova LA, Gnatovsky RY: Vertical mixing, size change and resting stage formation of the planktonic diatom Aulacoseira baicalensis. Eur J Phycol. 2010, 45 (4): 354-364. 10.1080/09670262.2010.492915.View ArticleGoogle Scholar
- Shimada C, Nakamachi M, Tanaka Y, Yamasaki M, Kuwata A: Effects of nutrients on diatom skeletal silicification: evidence from Neodenticula seminae culture experiments and morphometric analysis. Mar Micropaleontol. 2009, 73 (3–4): 164-177.View ArticleGoogle Scholar
- Cortese G, Gersonde R: Morphometric variability in the diatom Fragilariopsis kerguelensis: implications for Southern Ocean paleoceanography. Earth Planet Sc Lett. 2007, 257 (3–4): 526-544.View ArticleGoogle Scholar
- Cortese G, Gersonde R, Maschner K, Medley P: Glacial-interglacial size variability in the diatom Fragilariopsis kerguelensis: possible iron/dust controls?. Paleoceanography. 2012, 27: PA1208-View ArticleGoogle Scholar
- Marchetti A, Cassar N: Diatom elemental and morphological changes in response to iron limitation: a brief review with potential paleoceanographic applications. Geobiology. 2009, 7 (4): 419-431. 10.1111/j.1472-4669.2009.00207.x.View ArticlePubMedGoogle Scholar
- Shukla SKCX, Cortese G, Nayak GN: Climate mediated size variability of diatom Fragilariopsis kerguelensis in the Southern Ocean. Quaternary Sci Rev. 2013, 69: 49-58.View ArticleGoogle Scholar
- RaviKumar MS, Ramaiah N, Tang D: Morphometry and cell volumes of diatoms from a tropical estuary of India. Indian J Mar Sci. 2009, 38 (2): 160-165.Google Scholar
- du Buf H, Bayer MM: Automatic Diatom Identification. 2002, New Jersey, London, Singapore, Hong Kong: World Scientific Publishing Co. Pte. Ltd.Google Scholar
- Grima C, Tadeo F, Álvarez T, Arribas JL: Diatoms classification using frequency domain techniques. Jornadas de Automática; León. 2003, LeónGoogle Scholar
- Álvarez-Borrego J, Solorza S: Comparative analysis of several digital methods to recognize diatoms. Hidrobiológica. 2010, 20: 158-170.Google Scholar
- Luo Q, Gao Y, Luo J, Chen C, Liang J, Yang C: Automatic identification of diatoms with circular shape using texture analysis. J Softw. 2011, 6 (3): 428-435.View ArticleGoogle Scholar
- DIADIST: Diatom and desmid identification by shape and texture. [http://www.cs.cf.ac.uk/diadist/code.htm],
- ImageJ. http://imagej.nih.gov/ij/,
- Rosin PL: Measuring shape: ellipticity, rectangularity, and triangularity. Mach Vis Appl. 2003, 14 (3): 172-184. 10.1007/s00138-002-0118-6.View ArticleGoogle Scholar
- Mou DQ, Stoermer EF: Separating Tabellaria (Bacillariophyceae) shape groups based on fourier descriptors. J Phycol. 1992, 28 (3): 386-395. 10.1111/j.0022-3646.1992.00386.x.View ArticleGoogle Scholar
- Pappas J, Stoermer E: Fourier shape analysis and fuzzy measure shape group differentiation of Great Lakes Asterionella Hassall (Heterokontophyta, Bacillariophyceae). Proceedings of the 16th International Diatom Symposium. 2001, 485-501.Google Scholar
- Kermarrec L, Bouchez A, Rimet F, Humbert JF: First evidence of the existence of semi-cryptic species and of a phylogeographic structure in the Gomphonema parvulum (Kützing) Kützing Complex (Bacillariophyta). Protist. 2013, 164 (5): 686-705. 10.1016/j.protis.2013.07.005.View ArticlePubMedGoogle Scholar
- Falasco E, Blanco S, Bona F, Goma J, Hlubikova D, Novais MH, Hoffmann L, Ector L: Taxonomy, morphology and distribution of the Sellaphora stroemii complex (Bacillariophyceae). Fottea. 2009, 9 (2): 243-256.View ArticleGoogle Scholar
- Frankova M, Poulickova A, Neustupa J, Pichrtova M, Marvan P: Geometric morphometrics - a sensitive method to distinguish diatom morphospecies: a case study on the sympatric populations of Reimeria sinuata and Gomphonema tergestinum (Bacillariophyceae) from the River Becva Czech Republic. Nova Hedwigia. 2009, 88 (1–2): 81-95.View ArticleGoogle Scholar
- Vesela J, Neustupa J, Pichrtova M, Poulickova A: Morphometric study of Navicula morphospecies (Bacillariophyta) with respect to diatom life cycle. Fottea. 2009, 9 (2): 307-316.Google Scholar
- Vesela J, Urbankova P, Cerna K, Neustupa J: Ecological variation within traditional diatom morphospecies: diversity of Frustulia rhomboides sensu lato (Bacillariophyceae) in European freshwater habitats. Phycologia. 2012, 51 (5): 552-561. 10.2216/11-101.1.View ArticleGoogle Scholar
- Loke RE, Du Buf H: Identification by curvature of convex and concave segments. Automatic diatom identification. Edited by: du Buf H, Bayer MM. 2002, Singapore: World Scientific Publishing, 141-166.View ArticleGoogle Scholar
- Klingenberg CP: MorphoJ: an integrated software package for geometric morphometrics. Mol Ecol Resour. 2011, 11 (2): 353-357. 10.1111/j.1755-0998.2010.02924.x.View ArticlePubMedGoogle Scholar
- TPS series. http://life.bio.sunysb.edu/morph/,
- PAST. http://folk.uio.no/ohammer/past/,
- JMP. http://www.jmp.com/,
- The R Project for Statistical Computing. http://www.r-project.org/,
- SPSS. http://www-01.ibm.com/software/analytics/spss/,
- OpenCV (Open Source Computer Vision Library) Version 2.4.2. http://opencv.org/,
- NET wrapper for the OpenCV image processing library, Version 2.4.2. Emgu CV, a cross platform. http://www.emgu.com,
- Insight Segmentation and Registration Toolkit (ITK), Version 4.20. http://www.itk.org/,
- Download Microsoft .NET Framework 4 (Web Installer) from Official Microsoft Download Center. http://www.microsoft.com/en-us/download/details.aspx?id=17851,
- Download Microsoft Visual C++ 2010 SP1 Redistributable Package (x64) from Official Microsoft Download Center. 2010, http://www.microsoft.com/en-us/download/details.aspx?id=13523, SP1 Redistributable Package (x64) from Official Microsoft Download Center,
- ADIAC public image files. http://rbg-web2.rbge.org.uk/ADIAC/pubdat/downloads/public_images.htm,
- Otsu N: A threshold selection method from gray-level histograms. IEEE Trans Syst Man Cybern. 1979, 9 (1): 62-66.View ArticleGoogle Scholar
- Canny J: A computational approach to edge detection. IEEE Trans Pattern Anal Mach Intell. 1986, 30: 125-147.Google Scholar
- Lehmann G: Robust automatic threshold selection. Insight J. 2006, 2006: July - December. http://hdl.handle.net/1926/370.,Google Scholar
- Bradski G, Kaehler A: Learning OpenCV : Computer Vision with the OpenCV Library. 2008, Sebastopol: O’ReillyGoogle Scholar
- Sklansky J: Finding the convex hull of a simple polygon. Pattern Recognit Lett. 1982, 1 (2): 79-83. 10.1016/0167-8655(82)90016-2.View ArticleGoogle Scholar
- Gonzalez RC, Woods RE: Digital Image Processing. 2008, Prentice Hall: Upper Saddle River, New JerseyGoogle Scholar
- Claude J: Morphometrics with R. 2008, New York: Springer Science + Business Media, LLCGoogle Scholar
- Claude J: Morphometrics with R - Errata 1.81. 2010, Springer, http://www.isem.univ-montp2.fr/recherche/files/2012/01/Morphometrics_errata1.81.pdf,Google Scholar
- Kloster M: Digitale Bildsignalverarbeitung in der Bioinformatik: Methoden zur Segmentierung und Klassifizierung biologischer Merkmale am Beispiel ausgewählter Diatomeen. 2013, Emden: University of Applied Sciences Emden/LeerGoogle Scholar
- Nafe R, Schlote W: Methods of shape analysis of two-dimensional closed contours - a biologically important, but widely neglected field in histopathology. Electron J Pathol and Histol. 2002, 8 (2): 1-18.Google Scholar
- Hu M-K: Visual Pattern Recognition by Moment Invariants. In IRE Transactions on Information Theory. 1962, 8 (2): 179-187. 10.1109/TIT.1962.1057692.View ArticleGoogle Scholar
- Russ JC: The Image Processing Handbook. 2011, Boca Raton, London, New York: CRC Press, SixthGoogle Scholar
- Zunic J, Rosin PL: A Convexity Measurement for Polygons. Proceedings of the British Machine Vision Conference. 2002, Cardiff, UK: BMVC 2002Google Scholar
- Barber HG, Haworth EY: A guide to the morphology of the diatom frustule, Volume 44. 1981, Ambleside, Cumbria, UK: Freshwater Biological AssociationGoogle Scholar
- Zielinski U, Gersonde R, Sieger R, Fütterer D: Quaternary surface water temperature estimations: calibration of a diatom transfer function for the Southern Ocean. Paleoceanography. 1998, 13 (4): 365-383. 10.1029/98PA01320.View ArticleGoogle Scholar
- Algae World: Mann et al. 2004: images and morphometric data. http://rbg-web2.rbge.org.uk/algae/research/mann_etal_2004_data.html,
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.