SplineSeg: Spline Based Interactive Image Segmentation


We propose a novel framework using splines for learning driven figure-ground interactive segmentation. The task of interactive image segmentation, with user provided foreground-background labeled seeds or samples, is formulated as learning an interpolating pixel classification function that is then used to assign labels for all unlabeled pixels in the image. The spline function can be chosen from different classes to model the supervised pixel classifier which can use sparse scribbles seed points from the user and enabling a fast on-line implementation.

Gaussian Elastic Body Splines (GEBS) [1]

In our first and earliest work [1], we used the elastic body splines (EBS) which are recently introduced to capture tissue deformations within a physical model-based approach for non-rigid biomedical image registration. EBS model the displacement of points in a 3D homogeneous isotropic elastic body subject to forces. In particular, we utilized the Gaussian elastic body spline (GEBS) in our ICIP 2014 paper [1]. Experimental results demonstrate the applicability of the GEBS approach for natural image segmentation. The GEBS method for interactive foreground image labeling shows promise and outperforms a previous approach using the thin-plate spline model.

Segmentation Comparison with GEBS [1]

Result on a BSDS500 image

Input with Seeds

Ground Truth (BSDS500)

Our Result [1]

Thin Plate Spline

Multiquadric Splines (MQ) [4]

In our subsequent work [4], we considered multiquadric (MQ) splines. Experiment results on vascular network extraction shows promise and can be a valuable user assisted tool in biomedical imagery.

Segmentation Result with MQ [4]

Result on epi-fluorescence imagery - Note the highly accurate result on a very difficult segmentation problem


Application of SplineSeg in natural imagery:

[1] S. Meena, V. B. S. Prasath, K. Palaniappan, G. Seetharaman. Elastic body spline based image segmentation. IEEE International Conference on Image Processing (ICIP 2014), Paris, France, Oct 27-30, 2014. Proc. IEEE, pp. 4378-4382. doi:10.1109/ICIP.2014.7025888

Poster at figshare:10.6084/m9.figshare.1203638

Application of SplineSeg in aerial imagery:

[2] S. Meena, R. Pelapur, V. B. S. Prasath, K. Palaniappan, G. Seetharaman. Multiscale focus driven segmentation using elastic body splines. Geospatial Informatics, Fusion, and Motion Video Analytics VI, SPIE Defense+Security (DSS), Baltimore, MD, USA, 17 - 21 Apr 2016. Proc. SPIE.

[3] S. Meena, K. Palaniappan, G. Seetharaman, V. B. S. Prasath. Interactive target selection in aerial imagery using elastic body splines. Geospatial Informatics, Fusion, and Motion Video Analytics VI, SPIE Defense+Security (DSS), Baltimore, MD, USA, 17 - 21 Apr 2016. Proc. SPIE.

Application of SplineSeg in biomedical imagery:

[4] S. Meena, V. B. S. Prasath, Y. M. Kassim, R. J. Maude, O. V. Glinskii, V. V. Glinsky, V. H. Huxley, K. Palaniappan. Multiquadric spline-based interactive segmentation of vascular networks. 38th Annual International Conference EMBS (IEEE EMBS/EMBC), Orlando, USA, Aug 16-20, 2016. Proc. IEEE, pp. 5913-5916. doi:10.1109/EMBC.2016.7592074

Integrating interactive segmentation in online FireFly tool:

[5] U. Sampathkumar, V. B. S. Prasath, S. Meena, K. Palaniappan. Assisted ground truth generation using interactive segmentation on a visualization and annotation tool.. IEEE Applied Imagery Pattern Recognition (AIPR), Washington DC, USA, October 2016. Proc. IEEE. doi:10.1109/AIPR.2016.8010603

Poster at figshare:10.6084/m9.figshare.4036245

Comprehensive paper:

[6] ElasticMap: Sparse seed point-based label interpolation for interactive image segmentation. In preparation.

In preparation:

[7] Interactive spline based segmentation for refining globally convex active contours.

[8] Interactive image segmentation with inhomogeneous Navier partial differential equation.

[9] Fusing multiple features for interactive image segmentation.