Rostom Kachouri

Adaptive feature selection for heterogeneous image databases

Various visual characteristics based discriminative classification has become a standard technique for image recognition tasks in heterogeneous databases. Nevertheless, the encountered problem is the choice of the most relevant features depending on the considered image database content. In this aim, feature selection methods are used to remove the effect of the outlier features. Therefore, they allow to reduce the cost of extracting features and improve the classification accuracy. We propose, in this paper, an original feature selection method, that we call Adaptive Feature Selection (AFS). Proposed method combines Filter and Wrapper approaches. From an extracted feature set, AFS ensures a multiple learning of Support Vector Machine classifiers (SVM). Based on Fisher Linear Discrimination (FLD), it removes then redundant and irrelevant features automatically depending on their corresponding discrimination power. Using a large number of features, extensive experiments are performed on the heterogeneous COREL image database. A comparison with existing selection method is also provided. Results prove the efficiency and the robustness of the proposed AFS method.

Multi-model classification method in heterogeneous image databases

Automatic heterogeneous image recognition is a challenging research topic in computer vision. In fact, a general purpose images require multiple descriptors to cover their diverse category contents. However, not all extracted features are always relevant. Furthermore, simply concatenating multiple features may not be efficient for recognizing images in heterogeneous databases. In this context, we propose a new heterogeneous image recognition system, which aims to enhance the recognition results while decreasing the computational complexity. In particular, the proposed system is based on two complementary methods: adaptive relevant feature selection and multi-model classification method (MM-CM). Since it employs hierarchically selected features, the MM-CM ensures better classification accuracy and significantly less computation time than existing classification methods. The performance of the proposed image recognition system is assessed through two image databases and a large number of features. A comparison with competing algorithms from the literature is also provided. Our extensive experimental results show that an adaptive feature selection based MM-CM outperforms existing methods and improves the classification results in heterogeneous image databases.

Content description and classification for Image recognition system

In this paper, an heterogeneous image recognition system based on content description and classification is proposed. In this system and for an heterogeneous image database several features extraction methods are used and applied to better describes the images content. The features relevance is tested and improved through support vectors machines (SVMs) classifier of the consequent images index database. The obtained accuracy recognition of proposed system is evaluated on heterogeneous images database using precision/recall curves.

Feature extraction and relevance evaluation for heterogeneous image database recognition

Content-based image retrieval (CBIR) techniques are becoming increasingly important in various fields. One of the most important steps in CBIR systems is feature extraction. However, using not appropriate features in heterogeneous image database during retrieval process does not provide a complete description of an image. Indeed, each feature is able to describe some characteristics related to the shape, the color or the texture of the objects in image, but it can not cover the entire visual characteristics of the image. Therefore, many researchers have explored the use of multiple features to describe an image. In this paper, we propose the extraction and the relevance evaluation of several features for an heterogeneous image database classification and recognition, then we study the image retrieval system effectiveness with a new hierarchical feature model. The obtained results prove that using the new hierarchical feature model is more efficient than the use of the classical aggregated features in an image retrieval system.

Sélection adaptative de caractéristiques pertinentes et classification hiérarchique des images dans les bases hétérogènes

RÉSUMÉ. Dans les bases hétérogènes, les images appartiennent souvent à différentes classes thématiques et nécessitent une large description permettant leur reconnaissance. Cependant, les caractéristiques utilisées ne sont pas toujours adaptées au contenu de la base d’images considérée. Nous proposons dans cet article une nouvelle approche se basant sur deux originalités, à savoir la sélection adaptative de caractéristiques et la classification multimodèle intitulée MC-MM. La sélection adaptative permet de ne considérer que les caractéristiques les mieux adaptées au contenu de la base d’images utilisée. La méthode MCMM assure la reconnaissance des images en se servant hiérarchiquement des caractéristiques sélectionnées. Les résultats expérimentaux obtenus confirment l’efficacité et la robustesse de notre approche. ABSTRACT. In heterogeneous databases, images often provided from different sources and belong to different topics, hence there is a need for a large description to ensure efficient representation of their content. However, extracted features are not always adapted to the considered image database. In this paper we propose a new image recognition approach based on two innovations, namely adaptive feature selection and Multi-Model Classification Method (MC-MM). The adaptive selection considers only the most adapted features with the used image database content. The MC-MM method ensures image recognition using hierarchically selected features. Experimental results confirm the effectiveness and the robustness of our proposed approach. MOTS-CLÉS : extraction d’attributs, sélection adaptative des caractéristiques pertinentes, classification multi-modèle, reconnaissance d’images, bases hétérogènes.

Multiple kernel weighting based SVM for heterogeneous image recognition system

Kernel based machine learning such as Support Vector Machines (SVMs) have proven to be powerful for many database classification problems, especially for Content Based Image Retrieval systems (CBIR). Multiple Kernel Learning (MKL) approach was recently proposed to improve kernel based classification results. MKL approach depends essentially on the used kernels and the computation of the optimal weight coefficients. However in case of heterogeneous databases, the complexity to treat and classify images provides great difficultly to define and determine optimal kernel weights. We propose in this paper an original kernel weighting method, which is intended for Multiple Kernel based SVM classification. Depending on the relevance of kernel training rates, the proposed method allows us to ensure better classification accuracy and significantly less computation time.

Adaptive relevant feature selection and hierarchical image classification in heterogeneous databases

In heterogeneous databases, images often provided from different sources and belong to different topics, hence there is a need for a large description to ensure efficient representation of their content. However, extracted features are not always adapted to the considered image database. In this paper we propose a new image recognition approach based on two innovations, namely adaptive feature selection and Multi-Model Classification Method (MC-MM). The adaptive selection considers only the most adapted features with the used image database content. The MC-MM method ensures image recognition using hierarchically selected features. Experimental results confirm the effectiveness and the robustness of our proposed approach.