CBMI 2014 Program

Online Proceedings

CBMI 2014 Gold Sponsor

Back

CBMI Keynote

Can you trust your content based retrieval system?

Andreas Uhl, 09:15-10:30, Session Chair: Harald Kosch

Abstract:

Content-based media processing is not a domain well known for its high security-awareness. We will discuss questions related to security and privacy in such systems. On the one hand, content-based retrieval systems can be deceived by fraudulent manipulated queries, on the other hand data privacy is threatened especially in distributed search and retrieval setups. We will discuss eventual solutions as proposed in the context of biometric systems and will check for their generalisability to other content-based processing application areas.

Short Bio

Andreas Uhl is currently head of the Computer Sciences Department at the Paris Lodron University of Salzburg, Austria. There he heads the MultiMedia Signal Processing and Security (WaveLab) group, specialising in media security and watermarking, biometrics, medical image analysis, and image and video processing and compression topics in general. Andreas Uhl holds Master degrees in Pure Mathematics as well as Secondary/High School teacher education (with qualifications for mathematics, computer science, philosophy, and psychology), a PhD in applied mathematics, and is a tenured professor for computer science. He is an associate editor of ACM TOMCCAP, Signal Processing: Image Communication, EURASIP Journal of Image and Video Processing, and the ETRI Journal. In June 2014, he acts as a general chair for the 2nd ACM Workshop on Information Hiding and Multimedia Security 2014, to be held in Salzburg, Austria.

CBMI 2014 Klagenfurt

Back

CBMI Keynote

Video Browsing?

Klaus Schöffmann, 17:00-18:00, Session Chair: Laszlo Böszörmenyi

Abstract:

Video browsing tools combine content-based multimedia indexing with human-computer-interaction methods to provide interactive search in video content. Over the years many sophisticated tools have been proposed in the literature. These tools complement content-based video retrieval applications and are particularly useful in situations where no query can be formulated for a specific search need. Instead of relying only on content-based indexing methods they provide flexible interaction features and use abstract content visualization methods to support the interactive search process of the user. In this talk I will first motivate why video browsing is a powerful and convenient way of content-based search in video. I will discuss state-of-the-art tools designed for interactive search in video and outline how they integrate users in the search process and allow them to use their potential knowledge of the underlying content. I will also focus on ways to evaluate such tools, discuss the Video Browser Showdown competition and draw conclusions for further work on video browsing. The last part of the talk will focus on how to use modern features of mobile multimedia devices to develop novel video search tools.

Short Bio

Klaus Schoeffmann is senior assistant professor at the Institute of Information Technology at Klagenfurt University, Austria. His current research focuses on visual content analysis and interactive video search, including user interfaces, image/video content analysis, content visualization and interaction models for video search and retrieval. He is the author of numerous peer-reviewed publications on video browsing, video exploration, and video content analysis. He received his Ph.D. degree (Dr. techn.) from Klagenfurt University in June 2009. In his Ph.D. thesis he investigated possibilities to combine video browsing, video retrieval, and video summarization in order to allow for immediate video exploration. Klaus Schoeffmann teaches various courses in computer science (including Media Technology, Multimedia Systems, Operating Systems, and Distributed Systems) and he has (co-) organized international conferences, special sessions and workshops. He is member of the IEEE and the ACM and a regular reviewer for international conferences and journals in the field of Multimedia.

CBMI 2014 Klagenfurt

Back

Session

Content Based Retrieval

Session Chair: Savvas Chatzichristofis, Democritus University of Thrace, GR; Wed. 11:00-12:30

A Robust Audio Fingerprinting Method for Content-Based Copy Detection

Chahid Ouali, Pierre Dumouchel and Vishwa Gupta

This paper presents a novel audio fingerprinting method that is highly robust to a variety of audio distortions. It is based on unconventional audio fingerprints generation scheme. The robustness is achieved by generating different versions of the spectrogram matrix of the audio signal by using a threshold based on the average of the spectral values to prune this matrix. We transform each version of this pruned spectrogram matrix into a 2-D binary image. Multiple 2-D images suppress noise to a varying degree. This varying degree of noise suppression improves likelihood of one of the images matching a reference image. To speed up matching, we convert each image into an n-dimensional vector, and perform a nearest neighbor search based on this n-dimensional vector. We test this method on TRECVID 2010 content-based copy detection evaluation dataset. Experimental results show the effectiveness of such fingerprints even when the audio is distorted. We compare the proposed method to a state-of-the-art audio copy detection system. Results of this comparison show that our method achieves an improvement of 22% in localization accuracy, and lowers minimal normalized detection cost rate (min NDCR) by half for audio transformations T1 and T2.


Unsupervised Feature Learning for Content-based Histopathology Image Retrieval

Jorge A. Vanegas, John Arevalo and Fabio Gonzalez

This paper proposes a strategy for content-based image retrieval, which combines unsupervised feature learning (UFL) with the classical bag-of-features (BOF) representation. In BOF, patches are usually represented using standard classical descriptors (i.e., SIFT, SURF, DCT, among others). We propose to use UFL to learn the patch representation itself. This is achieved by applying a topographic UFL method, which automatically learns visual invariance properties of color, scale and rotation from an image collection. The learned image representation is used as input for a multimodal latent semantic indexing system, which enriches the visual representation with semantics from image annotations. The overall strategy is evaluated in a particular histopathology image collection retrieval task, showing that the learned representation has a positive impact in retrieval performance for this particular task.


Symmetry-Based Alignment for 3D Model Retrieval

Quoc Viet Dang, Sandrine Mouysset and Géraldine Morin

This paper presents an original method for aligning 3D shapes modeled as NURBS based B-rep models, exploiting partial symmetries of the shape. Aligning 3D shapes is an important pre-processing step in 3D shape retrieval and indexation: a reliable alignment of the shapes is necessary. Given a 3D model, three canonical planes define the normalized pose. We characterize the first canonical plane by an efficient computation of the dominant partial symmetry, using an efficient face matching approach. An area projection based algorithm determines the two relative remaining planes. To evaluate the robustness and the effectiveness, this technique is applied to 3D shape retrieval within a repository of NURBS based B-Rep models.


CBMI 2014 Klagenfurt

Back

Session

Large-Scale Multimedia Indexing and Search

Session Chair: Markus Schedl, Johannes Kepler University Linz, AT; Wed. 14:00-15:30

Using semantic features to improve large-scale visual concept detection

Mats Sjöberg and Jorma Laaksonen

Currently there are many multimedia benchmarks and databases available with a predefined set of concepts for which detectors can be formed or are even already available. One can use these background concepts to form semantic concept vectors by concatenating the concept prediction outputs for each image or video in the database. In this paper we investigate the use of such semantic concept features for detecting novel concepts in two large-scale experiments: the TRECVID 2012 evaluation with 800 hours of video data, and MIRFLICKR with 1 million images. We show that the detection performance can improve significantly over using visual features only. In some applications, computationally expensive kernel classifiers cannot be used in the detection phase, and our experiments show a consistent significant improvement using fast linear classifiers when we replace visual features with the semantic concept feature. We also propose a Self-Organising Map-based method which affords fast training-free detection and intuitive visualisation properties.


A large-scale audio and video fingerprints-generated database of TV repeated contents

Jean-Hugues Chenot and Gilles Daigneault

Using specifically-designed lightweight audio and video fingerprints, we have been able to detect repeated contents over a quasi-uninterrupted recording of 10+ TV channels, over more than 4 years, starting from January 2010 (380,000 hours); the detection is run using both audio and video fingerprints. The results are stored into a database that holds more than 20 million detected repeats. Detections range from a few seconds up to one hour. The database can be explored using a standard web browser. There are a number of potential applications, e.g. for structuring and documenting contents.


Fast Large-Scale Multimedia Indexing and Searching

Hisham Mohamed, Hasmik Osipyan and Stephane Marchand-Maillet

Searching for digital images in large-scale multimedia database is a hard problem due to the rapid increase of the digital assets. Metric Permutation Table is an efficient data structure for large-scale multimedia indexing. This data structure is based on the Permutation-based indexing, that aims to predict the proximity between elements encoding their location with respect to their surrounding. The main constraint of the Metric Permutation Table is the indexing time. With the exponential increase of multimedia data, parallel computation is needed. Opening the GPUs to general purpose computation allows to perform parallel computation on a powerful platform. In this paper, we propose efficient indexing and searching algorithms for the Metric Permutation Table using GPU and multi-core CPU. We study the performance and efficiency of our algorithms on large-scale datasets of millions of images. Experimental results show a decrease of the indexing time while preserving the quality of the results.


CBMI 2014 Klagenfurt

Back

Session

Special Session: Medical Image Analysis

Session Chair: Manfred del Fabro, Klagenfurt University, AT; Wed. 16:00-17:00

Evaluation of Super-Resolution Methods in the Context of Colonic Polyp Classification

Michael Häfner, Michael Liedlgruber, Andreas Uhl and Georg Wimmer

In this work we investigate whether it is possible to improve the results of an automated classification of colonic polyps by using super-resolution algorithms on endoscopic video sequences. For this purpose we apply different super-resolution methods to endoscopic sequences and use a set of feature extraction methods for the classification of the SR reconstruction results. We then compare the results obtained from these experiments against the classification results based on original low-resolution frames and against classification rates based on upscaled versions of low-resolution frames. We show that, at least for the set of super-resolution methods and feature extraction methods evaluated, applying super-resolution methods to the low-resolution frames has no significant impact on the resulting overall classification results.


Diffusion Tensor Imaging retrieval for Alzheimer’s disease detection

Ben Ahmed Olfa, Benois-Pineau Jenny, Allard Michelle, Catheline Gwenaelle and Ben Amar Chokri

Content-Based Visual Information Retrieval (CB- VIR) on MRI is penetrating the universe of IT tools sup- porting clinical decision making. A clinician can take profit from retrieving subjects’ scans with similar patterns. The CBIR approach has been used since recently for Alzheimer’s disease (AD) diagnosis. The most explored imaging modality in this context is the structural MRI. The Diffusion Tensor Imaging( DTI) is a relatively recent technique and CBIR approaches have not yet been developed on it. The combination of modalities in has to push the performances of CBIR methods to higher scores, but first of all it is necessary to explore the ability of DTI modality to give a correct answer on its own. The present research is the first attempt (in our best knowledge) to apply a CBIR technique on this modality for AD diagnosis. The proposed approach is based on the comparison of hippocampal areas. The degradation of them is a biomarker for AD. We use the Circular Harmonic Functions (CHFs) to extract content from the Diffusion Tensor -derived map : Mean Diffusivity (MD). This study was first accomplished with a subset of participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset and then tested on the DTI scans of French epidemiological study : ”Bordeaux dataset”. The results are promising and open interesting perspectives.


CBMI 2014 Klagenfurt

Back

Session

SessionName

Session Chair: xxx; time

title

authors


title

authors


title

authors


Online Proceedings

CBMI 2014 Gold Sponsor