Microfossils shape classification using a set of width values Roberto Marmo Dipartimento Informatica Sistemistica University of Pavia Via Ferrata 1 27100 Pavia, Italy marmo@vision.unipv.it Sabrina Amodio Istituto per l’Ambiente Marino Costiero National Research Council Calata Porta Di Massa, 80133 Napoli, Italy Virginio Cantoni Dipartimento Informatica Sistemistica University of Pavia Via Ferrata 1 27100 Pavia, Italy Abstract Recognition the shape of objects is a particularly rele- vant problem in pattern recognition. Foraminifera are very important microfossil to determine geological age of marine rocks. In this paper we propose an approach to shape recog- nition that takes as input the measurements of widths of the most common foraminifera shells after a pre-processing step to rotate the object in order to have vertical alignment. A k-NN and a MLP classifiers are compared for classifica- tion of chambers arrangement, experimental results show 87.1 and 97.1% of right answers, respectively. 1. Introduction Recognition of shapes is a fundamental task in computer vision, because shape is probably the most important prop- erty that is perceived about objects. The analysis consists in boundary extraction by segmentation technique of original grey scale image, then applying edge detectors and thin- ning algorithms to the binary image to retain the boundary width as small as possible. Once the boundary has been es- tablished the feature extraction task begins with extracting relevant features from each object. Foraminifera are single-celled organisms (protists) with minelarized shells [2, 6]. They are abundant and widespread microfossils (size 100μm - 20cm) in marine environments, moreover some species are short-lived. Foraminifera are very important to determine the geological age of the ma- rine rocks. Foraminifera shells are built with chambers which are added during growth and are separated by parti- tions with small openings called “foramina” connecting the chambers. They get name from these foramina [2, 6]. The common forms of chambers arrangement are (Fig. 1): 1. uniserial: chambers added in single linear series (A); 2. biserial: chambers added in double linear series (B); 3. milioline: chambers arranged in a series where each chamber extends the length and each successive cham- ber is placed at an angle of up to 180 degrees from the previous one (C); 4. spiral: chambers added in coil within single plane (D); Liu et al. [7] developed a knowledge-based system for the identification by computer vision of some representa- tives group of foraminifera. Schiebel et al. [9] developed a software for classification of six species of foraminifera on digital images, using neural network structures designed by operator. We use microfossils because they are characterized by many different outlier. We overcome the difficulties on de- scription of chambers arrangement by on a set of 11 width measurement, instead of considering the entire shape de- scription. The microfossils were examined with an optical micro- scope and acquired by digital camera with magnification about 20-25X, digitized with 150 dots per inch 256 grey levels. The image is converted in black-white 1-0 image: grey pixels become black and the white pixels are related to background. There is no holes in objects. In order to com- pare the outliers, the black object is rotated in a counter- clockwise direction, through degrees resulting from the dif- ference between the x-axis and the major axis in the carte- sian plane, so the major axis is vertical. Using the nearest 0-7695-2521-0/06/$20.00 (c) 2006 IEEE