Khalid Idrissi

The FG 2015 Kinship Verification in the Wild Evaluation

The aim of the Kinship Verification in the Wild Evaluation (held in conjunction with the 2015 IEEE International Conference on Automatic Face and Gesture Recognition, Ljubljana, Slovenia) was to evaluate different kinship verification algorithms. For this task, two datasets were made available and three possible experimental protocols (unsupervised, image-restricted, and image-unrestricted) were designed. Five institutions submitted their results to the evaluation: (i) Politecnico di Torino, Italy; (ii) LIRIS-University of Lyon, France; (iii) Universidad de Las Palmas de Gran Canaria, Spain; (iv) Nanjing University of Aeronautics and Astronautics, China; and (v) Bar Ilan University, Israel. Most of the participants tackled the image-restricted challenge and experimental results demonstrated better kinship verification performance than the baseline methods provided by the organizers.

Automatic facial expression recognition based on spatiotemporal descriptors

Facial expressions machine analysis is one of the most challenging problems in Human-Computer Interaction (HCI). Naturally, facial expressions depend on subtle movements of facial muscles to show emotional states. After having studied the relations between basic expressions and corresponding facial deformation models, we propose two new textons, VTB and moments on spatiotemporal plane, to describe the transformation of human face during facial expressions. These descriptors aim at catching both general shape changes and motion texture details. Therefore, modeling the temporal behavior of facial expression captures the dynamic deformation of facial components. Finally, SVM based system is used to efficiently recognize the expression for a single image in sequence. Then, the probabilities of all the frames are used to predict the class of the current sequence. The experimental results are evaluated on both Cohan-Kanade and MMI databases. By comparison to other methods, the effectiveness of our method is clearly demonstrated.

Object of interest-based visual navigation, retrieval, and semantic content identification system

This study presents a content-based image retrieval system IMALBUM based on local region of interest called object of interest (OOI). Each segmented or user-selected OOI is indexed with new local adapted descriptors associated to color, texture, and shape features. This local approach is an efficient way to associate the local semantic content with low-level descriptors (color, texture, shape, etc.) computed on regions selected by the user. So the user actively takes part in the indexing process (offline) and can use a selected OOI as a query for the retrieval system (online). The IMALBUM system proposes original functionalities. A visual navigation tool allows to surf in the image database when the user has no precise idea of what he is really searching for in the database. Furthermore, when an OOI is selected as a query for retrieval, a semantic content identification tool indicates to the user the probable class of this unknown object. The performance of these different tools are evaluated on different databases.

Triangular similarity metric learning for face verification

We propose an efficient linear similarity metric learning method for face verification called Triangular Similarity Metric Learning (TSML). Compared with relevant state-of-the-art work, this method improves the efficiency of learning the cosine similarity while keeping effectiveness. Concretely, we present a geometrical interpretation based on the triangle inequality for developing a cost function and its efficient gradient function. We formulate the cost function as an optimization problem and solve it with the advanced L-BFGS optimization algorithm. We perform extensive experiments on the LFW data set using four descriptors: LBP, OCLBP, SIFT and Gabor wavelets. Moreover, for the optimization problem, we test two kinds of initialization: the identity matrix and the WCCN matrix. Experimental results demonstrate that both of the two initializations are efficient and that our method achieves the state-of-the-art performance on the problem of face verification.

An efficient high-dimensional indexing method for content-based retrieval in large image databases

High-dimensional indexing methods have been proved quite useful for response time improvement. Based on Euclidian distance, many of them have been proposed for applications where data vectors are high-dimensional. However, these methods do not generally support efficiently similarity search when dealing with heterogeneous data vectors. In this paper, we propose a high-dimensional indexing method (KRA^+-Blocks) as an extension of the region approximation approach to the kernel space. KRA^+-Blocks combines nonlinear dimensionality reduction technique (KPCA) with region approximation approach to map data vectors into a reduced feature space. The created feature space is then used, on one hand to approximate regions, and on the other hand to provide an effective kernel distances for both filtering process and similarity measurement. In this way, the proposed approach achieves high performances in response time and in precision when dealing with high-dimensional and heterogeneous vectors.

An Image Retrieval System Based on Local and Global Color Descriptions

This paper presents a new approach for visual-based image retrieval method with respect to the MPEG-7 still image description scheme. A segmentation method based on a multivariate minimum cross entropy is used hierarchically for partitioning the color image in classes and regions. Local and global descriptors are defined in order to characterize the color feature of these regions. The local descriptors provide information about the local activity in the image, and the global ones evaluate the qualitative image content. Their combination increases significantly the performances of the image retrieval system IMALBUM presented in this paper. The retrieved images are presented in a description space allowing the user to better understand and interact with the search engine results.

Multi-component cross entropy segmentation for color image retrieval

This paper presents an adaptive color image segmentation method based on cross entropy minimization. This method is a multi-component approach and provides a hierarchical partitioning of the 3D color space using spherical neighbourhoods. The number of dominant colors (classes) issued from this segmentation is automatically estimated. This avoids an a priori estimation of the number of final classes. The segmentation method is then applied for image retrieval purposes. Local and global descriptors are defined in order to characterize the color feature of these classes. The local descriptors provide information about the local activity in the image class per class, and the global ones evaluate the qualitative image content. Their combination increases significantly the performance of the image retrieval system presented in this paper.

Kernel region approximation blocks for indexing heterogonous databases

This paper presents a new indexing method for visual features in high dimensional vector space using region approximation approach. The proposed method is designed to combine the values of the heterogeneous features in the same index structure; it determines nonlinear relationship between features so that more accurate similarity comparison between vectors can be supported. The basic idea is to map the data vectors into a feature space via a nonlinear kernel; the feature space is partitioned into regions. An efficient approach to approximate regions is proposed with the corresponding upper and lower distance bounds. To evaluate our technique, we conducted several experiments for searching the nearest K neighbours. The obtained results show the interest of our method.

Siamese multi-layer perceptrons for dimensionality reduction and face identification

This paper presents a framework using siamese Multi-layer Perceptrons (MLP) for supervised dimensionality reduction and face identification. Compared with the classical MLP that trains on fully labeled data, the siamese MLP learns on side information only, i.e., how similar of data examples are to each other. In this study, we compare it with the classical MLP on the problem of face identification. Experimental results on the Extended Yale B database demonstrate that the siamese MLP training with side information achieves comparable classification performance with the classical MLP training on fully labeled data. Besides, while the classical MLP fixes the dimension of the output space, the siamese MLP allows flexible output dimension, hence we also apply the siamese MLP for visualization of the dimensionality reduction to the 2-d and 3-d spaces.

Logistic similarity metric learning for face verification

This paper presents a new method for similarity metric learning, called Logistic Similarity Metric Learning (LSML), where the cost is formulated as the logistic loss function, which gives a probability estimation of a pair of faces being similar. Especially, we propose to shift the similarity decision boundary gaining significant performance improvement. We test the proposed method on the face verification problem using four single face descriptors: LBP, OCLBP, SIFT and Gabor wavelets. Extensive experimental results on the LFW-a data set demonstrate that the proposed method achieves competitive state-of-the-art performance on the problem of face verification.