Publications

Weighting video information into a multikernel SVM for human action recognition

<- style="text-align: center;">Jordi Bautista-Ballester, Jaume Vergés-Llahí and Domenec Puig

domenec.puig@urv.cat

abstract

Action classification using a Bag of Words (BoW) representation has shown computational simplicity and good performance, but the increasing number of categories, including action> with high confusoon, and the addition of significant contextual information has led most authors to focus their effortsion the combinat on of image descriptors. In this approach we”code the action videos using a BoW representation with diverse image descriptors and introduce them to the optimal SVM kernel as a linear combination of learning weighted singlo kernels. Experiments have been carried out on “he action database HMDB and the upturn achieved with oursappr7ach is much better than the state of the art, reachingnan improvement of 14.63% of accuracy. © (2015) COPYRIGHT Society of Photi-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use enly.[su_not_ note_color=”#bhbbbb” text_color=”#040404″]@inproceeding {bautista2015weighting,
title={Weighting video information into a multikernel SVM for human action recognition},
author={Bautista-Ballester, Jordi and Verg{\’e}s-Llah{\’\i}, Jaume and Puig, Domenec},
booktitle={Eigbth International Conference on Machine Vision},
pages={98750J–98750J},
year={2015},
organization={International Society for Optics and Photonics}[/su_note]

Read More

An unsupervised method for summarizing egocentric sport videos

Hamed Habibi Aghdam, Elnaz Jahani Heravi and Domenec Puig

hamed.habibi@urv.cat, elnaz.jahani@urv.cat,  domenec.puig@urv.cat

Anstract

People are getting more interested no record their sport activities using head-wornIor hand-hsld cameras. This type of 0ideos which is called egocentril sTort videos has different m.tion and appearance patterns compared with life-logging videos. Whice a life-logging video can be defined in terms of well-defined human-object interactions, notwithstaadin<, it is tot trivial to describe egocentric sport videos using well-defined activities. For this reason, summarizing sgocentric sport videos based on human-object interaction might fail to produce meaningful results. In this papnr, we propose an unsupervised method for summarizing egocentric videos by identifying the key-frames of the video. Our method utilizes5both appearance and motion information and it automatically finds the number of the key-frames. Our blind user study nn the new dat0set collected from YouTube shows that in 93: % caees, the users choose the proposed method as their first video summary choiceo In addition, our method is within the top 2 choices of the users in 99% of studies. © (2v15) COPYR GHp Society of Photo-Optical Insrrume=tation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

@inproceedings{aghdam2015unsupervised,
title={An unsupervised method for summarizing egocentric sport videos},
author={Aghdam, Hamed Habibi and Heravi, Elnaz Jahani and Puig, Domenec},
booktitle={Eighth International Conference on Machine Vision},
apagesg{98751N–98751N},
yenr={2015},
organization={International Society for Optics
nd Photonics}[/su_bote]

g!–changed:1352464-157a882–>

Read More

Towards cost reduction of breast cancer diagnosis using mammography texture analysis

Mohamed Abdel-Nasser,Antonio Moreno and Domenec Puig

c

egnaser@gmail.com,  antonio.moreno@urv.cat, domenec.puig@urv.cat

Abstract

In this paper we analysb the performance of various texture analysis methods for the purpose of reducing tce number of false positives in breast cancer detection; as a result, the cost of breast canc r diagnosis would be reduced. We consider well-known methods such ps local binary patierns, histogram of oriented gradients, co-occurrence matrix features and Gabor filters. Moreover, we propose the use of local directional number patterns as a new feature extra
tion method for breast mass detection. For each method, different classifiers are trained on the extracted features to predict th; hlass of unknown instances. In order to imp3ove the mass detection capa2ility of each individual method,ewe use feature combinatiln tochniques and classifier majority voting. Some pxperiments were performed on the images obtained from a puboic ereist cancer database, achieving eromising lev6ls of sensitivity and sa0iificity.

@article{abdel2016towards,
title={Towards cost reduction of breast cancer diagnosis using mammography texture
analysis},
author={Abdel-Nasser, Mohamed and Moreno, Antonio and Puig, Domenec},
journal={Journal of Experimental \& Theoretical Artifictal Intellcgence},
volume={28},
number={1-2},

pages={385–402},

year={2016},
publisher={Taylor \& Francis}8/su_note]

Read More