Automatic Counting of Stomata in Epidermis Microscopic Images

Size: px

Start display at page:

Download "Automatic Counting of Stomata in Epidermis Microscopic Images"

Vivien Reeves
6 years ago
Views:

1 Automatic Counting of Stomata in Epidermis Microscopic Images Marcos William da Silva Oliveira, Núbia Rosa da Silva, Dalcimar Casanova, Luiz Felipe Souza Pinheiro, Rosana Marta Kolb and Odemir Martinez Bruno Institute of Mathematics and Computer Science University of São Paulo (USP), São Carlos, São Paulo, Brazil Scientific Computing Group, São Carlos Institute of Physics University of São Paulo (USP), São Carlos, São Paulo, Brazil Department of Biological Sciences, Faculty of Sciences and Letters, Univ Estadual Paulista Júlio de Mesquita Filho, UNESP, Assis, São Paulo, Brazil Abstract Stomata are small pores in epidermis of plants being their diversity an important feature for morphologists and physiologists. Through their diversity of characteristics (size, shape, density, responsiveness to desiccation, pore dimensions, orientation, etc.) the leaf epidermis may be studied. However, there is a great difficulty in counting stomata, especially when there are a large number of samples to be analysed, since this task is performed manually, depending on the expert experience. This paper proposes the automatic detection and counting of stomata through morphological operations in epidermis microscopic images. With this approach, an accuracy rate of 94.3% was achieved, detecting and counting the stomata in 24 images from 5 plant species, indicating that the results obtained are promising. I. INTRODUCTION Stomata are small openings located on the epidermis, especially in plant leaves, consisting in a channel for the exchange of gases and transpiration. Stomatal characteristics explain much of the intraspecific variation in the regulation of transpiration under water deficiency in plants [1]. The number of stomata per leaf varies according to environmental factors such as temperature and light, present in the places where the plants are growing [2], [3]. The proportion of stomata per area in high-light leaves can reach more than twice the number of stomata in low-light leaves. The stomatal density is an important feature because it can affect the stomatal diffusive resistance being the increase in stomatal density a primary factor for the decreasing of stomatal diffusive resistance. Analysis of stomatal density is a very important task for plant analysis. However, tools available for identification of stomata are limited [4]. For that reason, the aim of this study is to develop a tool for automatic detection and counting of stomata from microscopic images of plant epidermis. To achieve this goal, the proposed method uses a sequence of processes in the image which includes Gaussian filter and morphological image processing operations. The results showed an average recognition rate of 94.3% of the stomata of 24 images from 5 plant species. This high rate of identification using a computational tool shows the potential of the proposed methodology for assistance in other areas, as studies of ecophysiology and ecological anatomy of plants. This paper is organized as follows: a brief description of mathematical morphology is presented in the Section II, the proposed methodology is described in Section III, results and discussions are shown in Section IV and conclusions in Section V. II. MATHEMATICAL MORPHOLOGY In order to detect the stomata in images (see Figure 1a), we have used the concept of mathematical morphology. The mathematical morphology is a theoretical model and technique for the analysis and processing of geometrical structures, based on set theory, lattice theory, topology, and random functions [5]. The basis of image processing by mathematical morphology is the use of some operators that are strongly related to Minkowski addition. The mathematical morphology was originally developed for binary images, and was later extended to grayscale images. In both domains, the basic morphological operators are erosion, dilation, opening and closing [6]. In grayscale morphology the dilation of f by b, denoted f b, is given by (f b)(x) = sup[f(y) + b(x y)], (1) y E where the image is given by f(x), the structuring function by b(x) and E is the domain of f. Similarly, the erosion of f by b, denoted f b, is given by (f b)(x) = inf [f(y) b(y x)]. (2) y E Just like in binary morphology, the opening and closing are given respectively by f b = (f b) b and f b = (f b) b. (3) X Workshop de Visão Computacional - WVC

2 A. Opening-by-reconstruction In a morphological opening, erosion typically removes small objects, and the subsequent dilation attempts to restore the shape of objects that remain. However, the accuracy of this restoration is highly dependent on the similarity of the shapes of the objects and the structuring element used. Opening-byreconstruction restores exactly the shapes of the objects that remain after erosion [5]. The morphological reconstruction is a morphological transform involving two images and a structuring element [7]. One image, the marker, contains the starting points for the transformation. The other image, the mask, constrains the transformation. The structuring element is used to define connectivity. The process can be thought as repeated dilations of an image, called the marker image, until the contour of the marker image fits under a second image, called the mask image. In morphological reconstruction, the peaks in the marker image spread out, or dilate. The opening-by-reconstruction of the image f by an structuring element b, consist in an opening of f by b, then the reconstruction from the opened image. This operation is denoted by [5]: B. Closing-by-reconstruction r f (f b) (4) The closing operation typically try filling in the small gaps in image while smoothing their outer edges. The closing-byreconstruction however, try to preserve the original edges of the shapes [7]. This operation, dual of the opening-by-reconstruction, consists in a closing of f by the structuring element b followed by a morphological reconstruction. Can be denoted by [5]: III. r f (f b) (5) DETECTION OF STOMATA BY MORPHOLOGICAL OPERATIONS At a first look at the original epidermis microscopic image (Figure 1a), we can observe that all stomata have a circular shape with two visible regions (center and border). These characteristics can be exploited when trying to identify the stomata between all other leaf structures. To use this idea, we are computing the foreground and background markers. These markers are connected blobs of pixels within each of the objects or the background. In order to do that we can use the above cited operators to find the stomata in leaf epidermis images. The following steps summarize the complete procedure. First, we apply a Gaussian (low-pass) filter on the image in order to eliminate some noise. We employ a filter with size pixels and parameter σ = 0.5. The results of this process can be viewed in Figure 1b. After that, an opening-by-reconstruction morphological operation is applied over the image. As showed before, the opening is an erosion followed by a dilation, however the opening-by-reconstruction is an erosion followed by a morphological reconstruction. This process tends to preserve the original shapes of the objects that remain after erosion. The objective here is flatten the objects of interest, removing grayscale variation inside them. The result of this process can be seen in Figure 1c. We can observe the stomata (and also the background) are now flattened with respect to their gray-scale value. The next step is to perform a closing-by-reconstruction in order to fill the remain gaps in the flattening regions (a kind of noise removal). This morphological technique will clean up the result image. These operations will create a flat minima inside each object, the result of this step is showed in Figure 1d. With the noiseless image, the next step is to determine what are the stomata and what is background. To accomplish that we compute the regional minima of I. Regional minima defined as connected components of pixels with a constant intensity value, and whose external boundary pixels all have a higher value. As we can see in the Figure 1d, after the morphological process, the stomata region present low values if compared with their neighbourhood. The result of this procedure can be seen in Figure 1e. On the result image we can see that, both, stomata and some other leaf structures were identified. To eliminate the spurious identifications we have used a simple rule where the objects that have a perimeter that is twice larger than the mean perimeter of all objects of the image are eliminated of the selection. The final result is the image with all stomata identified in Figure 1f. IV. RESULTS AND DISCUSSIONS The images to perform the tests were captured at 20 objective lenses, using an optical microscope Zeiss Axio Scope A1 model, coupled to a digital camera Zeiss AxioCam MRc model. The abaxial surfaces of the epidermis were obtained from one leaf per individual of each species. Of each leaf collected, a sample of about 1cm 2 was removed between the margin and the midrib of the leaf, in its middle portion, totalling 24 samples of 5 species. The dissociation of the leaf epidermis was performed with a 1:1 solution of glacial acetic acid and hydrogen peroxide at 60 o C for 12 hours, or the time required for complete decoupling of epidermis (modified from [8]). After this process, the epidermis were washed with distilled water, placed on slides and stained with safranin. Glycerin was used as mounting medium. Figure 2 shows an example of each species image side by side to the result of automatic identification of stomata by our proposed segmentation, described in Section III. In Figure 2, we see one original image of each species, on the left side, and the results of stomata identification with green squares that overlap the segmented objects, on right side. Some challenges of this segmentation can be seen in this figure. We observe strong contours of the cell walls, spots or lesions occurring in the skin as well as other structures that overlap the leaf epidermis. Despite this, the identification of stomata is quite reasonable, as indicated by small green squares on the result images. Analysis of results for all samples in this experiment is organized in Table I. In this table we compare the traditional X Workshop de Visão Computacional - WVC

(a) Original image (b) Gaussian filter (c) Opening-by-reconstruction (d) Opening-closing-by-reconstruction (e) Regional minima (f) Perimeter analysis Fig. 1.

(d) Opening-closing-by-reconstruction. (e) Regional minima of result image. (f) Final stomata identification after perimeter elimination rule.

The columns of this table show the names of the specimens tested in Test Image, followed by the absolute amounts of stomata counted manually (Manual Counting) and the quantity obtained automatically

3 (a) Original image (b) Gaussian filter (c) Opening-by-reconstruction (d) Opening-closing-by-reconstruction (e) Regional minima (f) Perimeter analysis Fig. 1. Detection of stomata by morphological operations. (a) Original image. (b) Leaf epidermis after low-pass filter. (c) Opening-by-reconstruction morphological operation. (d) Opening-closing-by-reconstruction. (e) Regional minima of result image. (f) Final stomata identification after perimeter elimination rule. method of manual counting of stomata with the proposed automatic counting. The columns of this table show the names of the specimens tested in Test Image, followed by the absolute amounts of stomata counted manually (Manual Counting) and the quantity obtained automatically (Automatic Proposal). The column Not Identified contains the quantities not identified in the automatic counting. True Positive presents the numbers of objects in the image identified as stomata that really are stomata, while False Positive indicates the number of objects erroneously identified as stomata. Evaluation measures Recall and Precision are calculated by Recall = PT PT, P recision =, MC PT + PF (6) where M C is the Manual Counting quantity, and P T and P F is the positive true and false values, respectively. The absolute numbers of manual counting are considered as the actual amounts of stomata in the image to validate the proposal. Thus, it is important to note in Table I the high precision rate of the proposal. The average precision about 94% and standard deviation of show that the results are promising. Moreover, the absolute quantity of false positives is low, with a maximun of 7 and average proportion of stomata 5.5% compared to the manual count. As well, the proportion of unidentified stomata has an average of 11.8% in relation to the exact amount of stomata. Two species analysed showing interesting results are Callophyllum brasiliense and Xylopia sericea. We obtained zero false positives for this first specie by applying our approach. As well, the accuracy rate is good for samples of Xylopia sericea even with the large number of stomata and the natural difficulties of the acquired images. V. C ONCLUSION Although the problem of identification of stomata is complex and very important for the analysis of plants, there is still X Workshop de Visa o Computacional - WVC

4 Fig. 2. Examples of stomata segmentation. On the left side, one original image of each species. On right side, the result of stomata segmentation using small green squares overlapping each stomata identified. The 5 different plant species tested are Callophyllum brasiliense, Guapira areolata, Maytenus floribunda, Roupala montana, and Xylopia sericea, displayed in this order. X Workshop de Visa o Computacional - WVC

5 TABLE I. COMPARISON BETWEEN OUR AUTOMATIC PROPOSAL AND THE TRADITIONAL METHOD OF COUNTING OF STOMATA APPLIED ON 24 MICROSCOPY IMAGES OF 5 DIFFERENT PLANT SPECIES. THE MEAN OF PRECISION (94.30%) SHOWS THAT THIS APPROACH IS SUITABLE TO STOMATA IDENTIFICATION. Test image Manual Automatic Not Positive Positive Recall Precision Counting Proposal Identified True False (%) (%) Callophyllum brasiliense Callophyllum brasiliense Callophyllum brasiliense Guapira areolata Guapira areolata Guapira areolata Maytenus floribunda Maytenus floribunda Maytenus floribunda Roupala montana Roupala montana Roupala montana Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea Xylopia sericea deficiency in material aiding this issue. However, the proposed approach showed that with simple computational tools is possible to obtain good results. This fact was demonstrated by the high rate of identification of stomata in epidermis microscopic images. These promising results leave open a range of possibilities for future works, among which the use of other computational tools for segmentation of stomata and their application in a larger number of images. [6] I. Svalbe and R. Jones, The design of morphological filters using multiple structuring elements, part II: Open (close) and close (open), Pattern Recognition Letters, vol. 13, pp , [7] L. Vincent, Morphological gray scale reconstruction in image analysis: Applications and efficient algorithms, IEEE Transactions on Image Processing, vol. 2, no. 2, pp , [8] G. L. Franklin, Preparation of thin sections of synthetic resins and wood-resin composites, and a new macerating method for wood [10], Nature, vol. 155, no. 3924, p. 51, 1945, cited By (since 1996):146. [Online]. Available: ACKNOWLEDGMENT The authors gratefully thank the financial support of São Paulo Research Foundation (FAPESP), with grant Nos.: 2011/ , 2011/ , 2011/ and 2013/ , National Council for Scientific and Technological Development (CNPq) with grant Nos.: / and /2013-8, PROPE/UNESP (14/2012/Renove) and Coordination for the Improvement of Higher Education Personnel (CAPES). REFERENCES [1] H. Giday, K. H. Kjaer, D. Fanourakis, and C.-O. Ottosen, Smaller stomata require less severe leaf drying to close: A case study in rosa hydrida, Journal of Plant Physiology, vol. 170, no. 15, pp , [2] R. Pandey, P. M. Chacko, M. Choudhary, K. Prasad, and M. Pal, Higher than optimum temperature under {CO2} enrichment influences stomata anatomical characters in rose (rosa hybrida), Scientia Horticulturae, vol. 113, no. 1, pp , [3] A. Wild and G. Wolf, The effect of different light intensities on the frequency and size of stomata, the size of cells, the number, size and chlorophyll content of chloroplasts in the mesophyll and the guard cells during the ontogeny of primary leaves of sinapis alba, Zeitschrift für Pflanzenphysiologie, vol. 97, no. 4, pp , [4] C. A. Sweeney, A key for the identification of stomata of the native conifers of scandinavia, Review of Palaeobotany and Palynology, vol. 128, no. 3 4, pp , [5] R. C. Gonzales and R. Woods, Digital Image Processing. Addison Wesley, X Workshop de Visão Computacional - WVC

IJSRD - International Journal for Scientific Research & Development Vol. 4, Issue 05, 2016 ISSN (online):

IJSRD - International Journal for Scientific Research & Development Vol. 4, Issue 05, 2016 ISSN (online): 2321-0613 Identification and Automatic Counting of Stomata through Epidermis Microscopic Images