Using extreme value theory for image detection

نویسندگان

  • Teddy Furon
  • Hervé Jégou
چکیده

The primary target of content based image retrieval is to return a list of images that are most similar to a query image. This is usually done by ordering the images based on a similarity score. In most state-of-the-art systems, the magnitude of this score is very different from a query to another. This prevents us from making a proper decision about the correctness of the returned images. This paper considers the applications where a confidence measurement is required, such as in copy detection or when a re-ranking stage is applied on a short-list such as geometrical verification. For this purpose, we formulate image search as an outlier detection problem, and propose a framework derived from extreme values theory. We translate the raw similarity score returned by the system into a relevance score related to the probability that a raw score deviates from the estimated model of scores of random images. The method produces a relevance score which is normalized in the sense that it is more consistent across queries. Experiments performed on several popular image retrieval benchmarks and state-of-the-art image representations show the interest of our approach. Key-words: image search, image detection, nearest neighbor, extreme value theory ha l-0 07 89 80 4, v er si on 1 18 F eb 2 01 3 Détection d’images par valeurs extrêmes Résumé : L’objectif de la recherche d’image est de retourner une liste d’images qui sont les plus visuellement similaires à une image requête, par exemple en ordonnant les images en fonction d’un score de similarité entre descripteurs d’images. Dans la plupart des systèmes de l’état de l’art, les scores varient significativement d’une requête à l’autre, ce qui empêche de déterminer la qualité intrinsèque des résultats retournés par la requête. Cet article considère des applications pour lesquelles une telle mesure de confiance est requise, comme en détection de copies ou pour le reclassement d’image avec un système de vérification géométrique. Dans cet objectif, nous formalisons le problème de la recherche d’image comme un problème de détection d’outlier, en utilisant le cadre mathématique de la théorie des valeurs extrêmes. Nous transformons le score de similarité en une probabilité que le score dévie du modèle de score des images quelconques. Cette probabilité est un nouveau score de pertinence qui est comparable d’une requête à l’autre, et utilisé comme une mesure de confiance. Des expériences effectuées sur des jeux d’évaluation usuels et plusieurs systèmes de recherche d’images montrent l’intérêt de notre approche. Mots-clés : recherche d’image, détection d’image, plus proches voisins, théorie des valeurs extrêmes ha l-0 07 89 80 4, v er si on 1 18 F eb 2 01 3 Using extreme value theory for image detection 3 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 20 40 60 80 100 120 140 160 180 200 di st an ce rank TP:query 1 TP:query 2 -0.25 -0.2 -0.15 -0.1 -0.05 0 0 20 40 60 80 100 120 140 160 180 200 no rm al iz ed s co re rank TP:query 1 TP:query 2 Figure 1: Illustration of the thresholding problem: Left: Two queries with the GIST descriptor in 1 million images. It is not possible to threshold the raw scores (here, Euclidean distances) to separate true positives from false positives. Right: The same queries, where our method have ‘normalized’ the raw scores into more meaningful relevance scores (log-likelihoods).

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

SIDF: A Novel Framework for Accurate Surgical Instrument Detection in Laparoscopic Video Frames

Background and Objectives: Identification of surgical instruments in laparoscopic video images has several biomedical applications. While several methods have been proposed for accurate detection of surgical instruments, the accuracy of these methods is still challenged high complexity of the laparoscopic video images. This paper introduces a Surgical Instrument Detection Framework (SIDF) for a...

متن کامل

Integration of Visible Image and LIDAR Altimetric Data for Semi-Automatic Detection and Measuring the Boundari of Features

This paper presents a new method for detecting the features using LiDAR data and visible images. The proposed features detection algorithm has the lowest dependency on region and the type of sensor used for imaging, and about any input LiDAR and image data, including visible bands (red, green and blue) with high spatial resolution, identify features with acceptable accuracy. In the proposed app...

متن کامل

Noisy images edge detection: Ant colony optimization algorithm

The edges of an image define the image boundary. When the image is noisy, it does not become easy to identify the edges. Therefore, a method requests to be developed that can identify edges clearly in a noisy image. Many methods have been proposed earlier using filters, transforms and wavelets with Ant colony optimization (ACO) that detect edges. We here used ACO for edge detection of noisy ima...

متن کامل

Frequency-Dependent Peak-Over-Threshold algorithm for fault detection in the spectral domain

An original novelty detection algorithm in the Fourier domain, using extreme value theory (EVT) is considered in this article. Periodograms may be considered as frequency-dependent random variables, and this can be taken into account when designing statistical tests. Frequency-Dependent Peak-Over-Threshold (FDPOT) puts special emphasis on the frequency dependence of extreme value statistics, th...

متن کامل

Poster: An Extreme Value Theory Approach to Anomaly Detection (EVT-AD)

We introduce a new approach to anomaly detection based on extreme value theory statistics. Our method improves detection accuracy by replacing binary feature thresholds with anomaly scores and by modeling the tail region of the distribution where anomalies occur. It requires no optimization or tuning and provides insights into results. This work describes the Extreme Value Theory-Anomaly Detect...

متن کامل

Novelty Detection using Extreme Value StatisticsStephen

Extreme value theory (EVT) is a branch of statistics which concerns the distributions of data of unusually low or high value i.e. in the tails of some distribution. These extremal points are important in many applications as they represent the outlying regions of normal events against which we may wish to deene abnormal events. In the context of density modelling, novelty detection, or Radial-B...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2013