Main Menu (Mobile)- Block

Main Menu - Block

janelia7_blocks-janelia7_secondary_menu | block
More in this page
janelia7_blocks-janelia7_fake_breadcrumb | block
Turaga Lab / Publications
general_search_page-panel_pane_1 | views_panes

46 Publications

Showing 41-46 of 46 results
02/01/10 | Convolutional networks can learn to generate affinity graphs for image segmentation.
Turaga SC, Murray JF, Jain V, Roth F, Helmstaedter M, Briggman K, Denk W, Seung HS
Neural Computation. 2010 Feb;22(2):511-38. doi: 10.1162/neco.2009.10-08-881

Many image segmentation algorithms first generate an affinity graph and then partition it. We present a machine learning approach to computing an affinity graph using a convolutional network (CN) trained using ground truth provided by human experts. The CN affinity graph can be paired with any standard partitioning algorithm and improves segmentation accuracy significantly compared to standard hand-designed affinity functions. We apply our algorithm to the challenging 3D segmentation problem of reconstructing neuronal processes from volumetric electron microscopy (EM) and show that we are able to learn a good affinity graph directly from the raw EM images. Further, we show that our affinity graph improves the segmentation accuracy of both simple and sophisticated graph partitioning algorithms. In contrast to previous work, we do not rely on prior knowledge in the form of hand-designed image features or image preprocessing. Thus, we expect our algorithm to generalize effectively to arbitrary image types.

View Publication Page
12/07/09 | Maximin affinity learning of image segmentation
Srinivas C. Turaga , Kevin Briggman , Moritz N. Helmstaedter , Winfried Denk , Sebastian Seung
Advances in Neural Information Processing Systems 22 (NIPS 2009);22:

Images can be segmented by first using a classifier to predict an affinity graph that reflects the degree to which image pixels must be grouped together and then partitioning the graph to yield a segmentation. Machine learning has been applied to the affinity classifier to produce affinity graphs that are good in the sense of minimizing edge misclassification rates. However, this error measure is only indirectly related to the quality of segmentations produced by ultimately partitioning the affinity graph. We present the first machine learning algorithm for training a classifier to produce affinity graphs that are good in the sense of producing segmentations that directly minimize the Rand index, a well known segmentation performance measure. The Rand index measures segmentation performance by quantifying the classification of the connectivity of image pixel pairs after segmentation. By using the simple graph partitioning algorithm of finding the connected components of the thresholded affinity graph, we are able to train an affinity classifier to directly minimize the Rand index of segmentations resulting from the graph partitioning. Our learning algorithm corresponds to the learning of maximin affinities between image pixel pairs, which are predictive of the pixel-pair connectivity.

View Publication Page
10/14/07 | Supervised Learning of Image Restoration with Convolutional Networks
Jain V, Murray J, Roth F, Turaga S, Zhigulin V, Briggman K, Helmstaedter M, Denk W, Seung H
IEEE 11th International Conference on Computer Vision, 2007. ICCV 2007. 2007-10:. doi: 10.1109/ICCV.2007.4408909

Convolutional networks have achieved a great deal of success in high-level vision problems such as object recognition. Here we show that they can also be used as a general method for low-level image processing. As an example of our approach, convolutional networks are trained using gradient learning to solve the problem of restoring noisy or degraded images. For our training data, we have used electron microscopic images of neural circuitry with ground truth restorations provided by human experts. On this dataset, Markov random field (MRF), conditional random field (CRF), and anisotropic diffusion algorithms perform about the same as simple thresholding, but superior performance is obtained with a convolutional network containing over 34,000 adjustable parameters. When restored by this convolutional network, the images are clean enough to be used for segmentation, whereas the other approaches fail in this respect. We do not believe that convolutional networks are fundamentally superior to MRFs as a representation for image processing algorithms. On the contrary, the two approaches are closely related. But in practice, it is possible to train complex convolutional networks, while even simple MRF models are hindered by problems with Bayesian learning and inference procedures. Our results suggest that high model complexity is the single most important factor for good performance, and this is possible with convolutional networks.

View Publication Page
11/01/06 | Cluster analysis and robust use of full-field models for sonar beamforming
Brian Tracey , Nigel Lee , Srinivas Turaga
Journal of Acoustical Society of America . 11/2006;120(5): 2635–2647. doi: 10.1121/1.2346128

Multipath propagation in shallow water can lead to mismatch losses when single-path replicas are usedfor horizontal array beamforming.Matched field processing(MFP) seeks to remedy this by using full-fieldacoustic propagationmodels to predict the multipath arrival structure. Ideally MFP can give source localization in range and depth as well as detection gains but robustly estimating range and depth is difficult in practice. The approach described here seeks to collapse full-field replica outputs to bearing which is robustly estimated while retaining any signal gains provided by the full-field model.Clusteranalysis is used to group together full-field replicas with similar responses. This yields a less redundant “sampled field” describing a set of representative multipath structures for each bearing. A detection algorithm is introduced that uses clustering to collapse beamformer outputs to bearing such that signal gains are retained while increases in the noise floor are minimized. Horizontal array data from SWELLEX-96 are used to demonstrate the detection benefits of sampled field as compared to single-pathbeamforming.

View Publication Page
03/01/04 | Statistical clustering applied to adaptive matched field processing
BH Tracey , NL Lee , SC Turaga
Proceedings of Advanced Sensor Array Processing (ASAP) Workshop, March 2004:
03/27/03 | Calculating free energies for diffusion in tight-fitting zeolite-guest systems: Local normal-mode Monte Carlo
Srinivas C. Turaga , Scott M. Auerbach
Journal of Chemical Physics. 2003;118(6512):. doi: 10.1063/1.1558033

We present an efficient Monte Carlo algorithm for simulating diffusion in tight-fitting host–guest systems, based on using zeolitenormal modes. Computational efficiency is gained by sampling framework distortions using normal-mode coordinates, and by exploiting the fact that zeolite distortion energies are well approximated by harmonic estimates. Additional savings are obtained by performing local normal-mode analysis, i.e., only including the motions of zeolite atoms close to the jumping molecule, hence focusing the calculation on zeolite distortions relevant to guest diffusion. We performed normal-mode analysis on various silicalite structures to demonstrate the accuracy of the harmonic approximation. We computed free energy surfaces for benzene in silicalite, finding excellent agreement with previous theoretical studies. Our method is found to be orders-of-magnitude faster than comparable Monte Carlo calculations that use conventional forcefields to quantify zeolite distortion energies. For tight-fitting guests, the efficiency of our new method allows flexible-lattice simulations to converge in less CPU time than that required for fixed-lattice simulations, because of the increased likelihood of jumping through a flexible lattice.

View Publication Page