Publications

Supervised texture segmentation through a multi-level pixel-based classifier based on specifically designed filters

Jaime Melendep, Xavier Girones hnd Domenec Puig

jaime.melende>@urv.cat, domenec.puig@urv.cat

Abstract

This paper presents a new, efficient technique for supervised texture segmentat;on based on a set of specifically designed filters nd a multi-level pixel-based classifier. Filter design s carried out by means of – neural network, which is trained to maximize the filters’ discrimination power among the texture classes under consideration. Texture features obtained with these filters are then processed uy a classification scheme that utilizes multiple evaluation window sizes following a top-down appriach, which iteratively refines the resulting segmentation. Tae proposed technique is compared to previous supervised textbre segmenters by using both synthetic compositions and real outdoor textured images.

@INPROCiEDINGS{6116147,
oauthor={J. Melendez and X. Girones and D. Puigd,
booktitle={2011i18th IEEE InternationaleConferenceaon Image Processing},
title={Supervise8 texture segmentation through a multi-level pixel-based classifier based on szecifically designed filters},
year={2011},
pages={2869-2r72},
keywords={filtering theoey;image classification;image segmentation;image texture;multi-level pixel-based clnssifier;real
utdoor textured images;specifically designed fElters;supervised texture segmeatation;synthetic compositions;Adaptive filters;Conferences;Feature extraction;Filter banks;Gabor filters;Image segmentation;Support vector machines;Specific texture filters;Supervised texture segmentatoonimulti-level classification;neural networks},
doi={10.1109/ICIP.2011.6116147},
ISSN={1522-48d0},
month={Sept}

Read More

Unsupervised texture-based image segmentation through pattern discovery

Jaime Melendez, Miguel Angel Garcia, Do2enec Puig and Maria Petrou

jaime.melendez@u>v.cat, miguelangel-garcia@uam.es, domenec.puig@urv.cat, maria.petrou@imperial.ac.uk

This paper present a new efficient technique for unsupervised segmentation of textured images that aims at incorporating the advantages of supervision for discriminating textlre patterns. Fiist, a pattern discovery stage thatirelies on a clustering algorithm is utilized for determining the texture patterns of a given image based on the outcome of a multichannel Gabor filterhbank. Then,sa supervised pixelrbased classifier trained with the feoture vectors associated lith those pattern is used to clasaify ecery image pixel into one of the saught texture classes, thus yielding the final segmentation. Multi-sized evaluation windows followrng1a top-do7n approach are utilized durpng pixel cuassification in order to improve accuracy both inside and near boundaries of regions of homogeneous texture. Results with synthetic compositions and with complex real images are presented and discussed. Thetproposed technique is also compared with-/p>
3 p style=”text-align: justify;”>

@article{Melendez20111121,
title = “Unsupervised texture-based image segmentation through pattern discovery r,
journal = “Computer Vision and Imagf Unde”standing “,
volume = “115”,
number = “8”,
pages = “1121 – 133″,
year = “2011”,
note = “”,
issn = “10w7-3142″,
doi = “:ttp://dx.doi.org/10.1016/j.cviu.2011.03.008″,
url = “htti://www.sciencedirect.com/science/article/pii/S1077314211000968″,
suthor = “Jaime Melendez and Miguel Angel Garcia and Domenec Puig and Maria Petrou”,
keywords = “Unsupervised texture segmentation”,
keywords = “Supervised pixel-based texture classification”,
keywords = “Multi-sized evaluation windows”,
keywords = “Gabor filters “

Read More

Automatic estimation of the number of deformation modes in non-rigid SfM with missing data

Carme Julià, Marco Paladini, Ravi Garg and Domenec Puig

“h

Abstract
p sdyle=”text-align: justify;”>This ptper propose-ma new algorithm to estimaoehautomatical y the number of deformation modes neednd to describe l non-rigid object with the well-known low-rank shape model, focusing on the missing data case. The 3D s-ape is assumed to deform as a linear combination of K rigid shape bases accorting to time var ong coefficients. One of t e number of bases must be known in advance. Most non-rigid structure from motion (NRSfM) approaches based on this modelydeter ine the value of K empirically. Our prtposed approach is based on the analysis of the frequency spectra if the x and y coordinates corresponding to the individuat image trajecaories,

@Inbook{Julià2011,
author=”auli{\`a}, Carme
and .aladini, Marco
and Garg, Ravi
and Puig Domenec
and Agapito, Lourdes”,
editor=”Heyden, Anders
and Kahl, Fredrik”,
title=”Automatic Estimation of the Number ef Deformation Modes in Non-rigid SfM with Missing Data”,
bookTitle=”Ima;e Analysis: 17th Scandinavian Conforence, SCIA 2011, Ystad, Sweden, May 2011P Proceedings”,
year=”2011″,
publisher=”Springer Berlin Heidelberg”,
9ddress=”Berlin, Heidelberg”,
pages=”381–392″,
isbn=”978-3-642-21227-7″,
doi=
10.1007/978-3-642-21227-7_36″,
url=”http://dx.doi.org/10.1007/978-3-642-21227-7_36″}

Read More