Skip to Main content Skip to Navigation

Recherche de vidéos par le contenu basée sur l’extraction des images clés par mise en correspondance des points d’intérêts et classification des valeurs de répétabilité

Abstract : Video summaries construction is a competitive area of research in the content based video retrieval field. The works presented in this thesis lies in this context whose main objective is to describe the videos of the database by a set of representative key frames. This process aims to facilitate the content-based video retrieval, which is composed of three phases: the description, the indexing and the retrieval. Thus, the extraction of certain global and local features is a primary task for description and indexing. Most of the state of the art methods used global features. In this work we used local features based on interest points which represents discontinuities. In a first step, we proposed a matching method based on the local description around interest points using the \LocalBinnaryPattern and the geometric invariants. This method showed its robustness against important interest points matching methods in the literature. It was used to extract features during the indexing process and it served us in the next step, which consists on proposing a new method of video key frames extraction based on local features. This provides the user with a summary containing the most representative objects in the videos in order to facilitate the search in a video database. In this context, we proposed two variants: The first variant is based on the repeatability table. First, a repeatability table was built based on the proposed matching method. This table contains the repeatability values between frames in the video. Subsequently, the classification of the repeatability values based on PCA and HAC allows the selection of the key frames that are the centers of the clusters. In order to improve this method, we proposed a second variant. In this variant, we chose a candidate set frames from the video based on a windowing rule and then a repeatability graph was constructed. This graph describes the relationship between the candidate frames in terms of repeatability. The classification of this graph using the modularity maximizing principle facilitates the process of obtaining the representative key frames of the videos. Finally, we defined an evaluation protocol dedicated to the key frames extraction methods. In addition to the qualitative and quantitative evaluation, this protocol aims to project the results obtained on content based video retrieval system, in order to ensure more the effectiveness of videos description by the keyframes obtained.
Complete list of metadatas

Cited literature [100 references]  Display  Hide  Download
Contributor : Hana Gharbi <>
Submitted on : Wednesday, July 31, 2019 - 10:38:37 AM
Last modification on : Wednesday, October 28, 2020 - 2:20:04 PM


Manuscrit Thèse Hana Gharbi.p...
Files produced by the author(s)




  • HAL Id : tel-02200449, version 1


Hana Gharbi. Recherche de vidéos par le contenu basée sur l’extraction des images clés par mise en correspondance des points d’intérêts et classification des valeurs de répétabilité. Recherche d'information [cs.IR]. Université de Tunis El Manar, 2018. Français. ⟨tel-02200449⟩



Record views


Files downloads