Boosting performance of the edge-based active contour model applied to phytoplankton images Adas Gelzinis * , Evaldas Vaiciukynas * , Marija Bacauskiene * , Antanas Verikas *† , Sigitas Sulcius ‡ , Ricardas Paskauskas ‡ and Irina Olenina ‡§ * Department of Electrical & Control Equipment, Kaunas University of Technology, Studentu 50, LT-51368 Kaunas, Lithuania † Intelligent Systems Laboratory, Halmstad University, Box 823, S-30118 Halmstad, Sweden ‡ Coastal Research and Planning Institute, Klaipeda University, Herkaus Manto 84, LT-92294 Klaipeda, Lithuania § Department of Marine Research, Environmental Protection Agency, Taikos Av. 26, LT-91144 Klaipeda, Lithuania Abstract—Automated contour detection for objects represent- ing the Prorocentrum minimum (P. minimum) species in phy- toplankton images is the core goal of this study. The species is known to cause harmful blooms in many estuarine and coastal environments. Active contour model (ACM)-based image segmentation is the approach adopted here as a potential solution. Currently, the main research in ACM area is highly focused on development of various energy functions having some physical intuition. This work, by contrast, advocates the idea of rich and diverse image preprocessing before segmentation. Advantage of the proposed preprocessing is demonstrated experimentally by comparing it to the six well known active contour techniques applied to the cell segmentation in microscopy imagery task. Index Terms—Active contour model, Energy function, Contour detection, Image segmentation, Image preprocessing, Phytoplank- ton images I. I NTRODUCTION Active contour models (ACMs) have been widely used for image segmentation [1]–[5], analysis of medical images [6], analysis of protein spots in two-dimensional gel electrophore- sis images and is one of the most successful techniques aiding object detection in various domains. ACM is based on solving energy minimization problem with incorporated a priori knowledge. To extract the contour of an object, the ACM seeks to evolve a curve which, under some predefined constrains, fits the object best. ACM is a curve c(s)=[x(s),y(s)],s ∈ [0, 1] deforming on the image region, to attain the desired properties. Curve deformation is achieved by minimizing such energy function: E(c)=  1 0  1 2 ( α‖c ′ (s)‖ 2 + β‖c ′′ (s)‖ 2 ) + E ext [c(s),f (I )]  ds (1) where c ′ (s) and c ′′ (s) are the first and the second derivative of c(s) with respect to s, α and β parameters, namely membrane term and thin plate term, control the sensitivity with respect to the first and the second derivative, and E ext corresponds to the external energy, linking the contour c(s) to specific features f (I ) of the image I . Both static (depending on the image) and dynamic external forces (depending on the contour) have been used. Static forces can be evaluated using image region intensity and texture information [7]. The following two ex- ternal energy functionals are typical for placing a contour on the edges [1]: E ext (x, y)= −‖∇I (x, y)‖ 2 (2) E ext (x, y)= −‖∇[G σ (x, y) ∗ I (x, y)]‖ 2 (3) where ∇ is the gradient operator, G σ (x, y) is a two- dimensional Gaussian of standard deviation σ, and ∗ stands for the convolution operation. Edge-based [1], [8] and region-based [2], [9] models are distinguished by the constraints applied. Edge-based models suffer from nearby adjacent objects and strong edges inside an object of interest, since these act as an attraction source for the active contour. Region-based models use statistical information of inside and outside parts of the contour and, in images with weak edges, tend to outperform edge-based solutions. ACM proposed by Chan and Vese (C-V method) is a widely accepted representative of region-based models [2]. The C-V method as well as some other popular ACM approaches assume constant intensity in various image regions [2]. More advanced techniques attempt to model regions by known distributions, intensity histograms, texture maps, or structure tensors [10]. Sadeghi et al. [11] employed the self-organizing map (SOM) to improve ACM. Zhang et al. combined the C-V method with geodesic active contour (GAC [8]) framework [12]. To enable detec- tion of partially occluded objects, Fang and Chan suggested incorporating shape prior model into the GAC equation [13]. Collection of different shapes was used as training data. Liu et al. introduced the shape prior constraint to guide the evolution of active contour [14]. Brox and Cremers [15] suggested incorporating local statistics into a variational framework. Local image information has also been used by Zhang et al. [16], to enable segmentation of images with intensity inho- mogeneities. Paragios and Deriche [17] suggested minimizing a sum of edge-based and region-based energies, while Sum and Cheung [18] minimized the sum of global and local energy derived from image contrast. Truc et al. [6], aiming to minimize object inhomogeneity and to maximize the distance between the object and the background, have also combined CINTI 2012 • 13th IEEE International Symposium on Computational Intelligence and Informatics • 20–22 November, 2012 • Budapest, Hungary 273 978-1-4673-5206-2/12/$31.00 ©2012 IEEE