Mustapha Kardouchi

Automatic birdsong recognition based on autoregressive time-delay neural networks

A template-based technique for automatic recognition of birdsong syllables is presented. This technique combines time delay neural networks (TDNNs) with an autoregressive (AR) version of the backpropagation algorithm in order to improve the accuracy of bird species identification. The proposed neural network structure (AR-TDNN) has the advantage of dealing with a pattern classification of syllable alphabet and also of capturing the temporal structure of birdsong. We choose to carry out trials on song patterns obtained from sixteen species living in New Brunswick province of Canada. The results show that the proposed AR-TDNN system achieves a highly recognition rate compared to the baseline backpropagation-based system

Improving Bag of Visual Words Image Retrieval: A Fuzzy Weighting Scheme for Efficient Indexation

Recent works on Content Based Image Retrieval rely on Bag of Visual Words to index images. Analogically to the Bag of Words approach used in text retrieval, this model allows describing an image as a bag of elementary local features called visual words. As a result, an image is represented by a vector of weights, where each weight corresponds to the importance of a visual word in the image. The choice of local features and the weighting scheme are very important to perform image retrieval. The existing weighting schemes are mostly migrated from text retrieval domain and don’t take into account fundamental differences between textual words and visual words. In this paper, a novel approach based on Scale Invariant Features Transform (SIFT) features and a new weighting scheme is proposed. The proposed scheme uses a fuzzy representation to index images with a more robust signature. Experimental results with the Coil-100 image database demonstrate that the proposed method produces better performance than known term weighting representations.

Face Recognition System Using Gabor Features and HTK Toolkit

This paper presents a new face recognition system. The proposed system is built on Hidden Markov Models (HMMs). Facial image features are extracted using Gabor filters. The dimensionality of those features is reduced using the Linear Discriminant Analysis (LDA) method to keep only the most relevant information. Then, the system injects the resulting feature vectors to the Hidden Markov Model Toolkit (HTK). Note that HTK is a portable toolkit for speech recognition system. Experimental results on YALE and ORL databases show the efficiency of the proposed approach.

Fuzzy indexing for Bag of Features scene categorization

This paper presents a novel Bag of Features (BoF) method for image classification. The BoF approach describes an image as a set of local descriptors using a histogram, where each bin represents the importance of a visual word. This indexing approach has been frequently used for image classification, and we have seen several implementations, but crucial representation choices — such as the weighting schemes — have not been thoroughly studied in the literature. In our work, we propose a Fuzzy model as an alternative to known weighting schemes in order to create more representative image signatures. Furthermore, we use the Fuzzy signatures to train the Gaussian Naïve Bayesian Network and classify images. Experiments with Corel-1000 dataset demonstrate that our method outperforms the known implementations.

Histograms of fuzzy oriented gradients for face recognition

Efficient face descriptors require a careful equilibration between accuracy and feature dimension. In recent years Histogram of Oriented Gradient (HOG) starts to be used in the face recognition task. However the best recognition rate for HOG requires a high dimensional feature. In this paper, we will incorporate fuzzy concept to HOG aiming to achieve a good recognition rate with a low feature vector dimension. The proposed Histogram of Fuzzy Oriented Gradient is applied to the face recognition task. Experimental results on ORL database have demonstrated that HFOG outperforms the original HOG with a lower dimensional vector.

Rotation invariant fuzzy shape contexts based on eigenshapes and fourier transforms for efficient radiological image retrieval

This paper proposes a new descriptor for radiological image retrieval. The proposed approach is based on fuzzy shape contexts, Fourier transforms and Eigenshapes. At first, fuzzy shape context histograms are computed. Then, a 2D FFT is performed on each 2D histogram to achieve rotation invariance. Next, histograms are projected onto a lower dimensionality feature space. The new space is more representative. It highlights the most important variations between shapes. Eigenshapes are the principal components for radiological images. The proposed approach is translation, scale and rotation invariant. Classes of the medical IRMA database are used for experiments. Comparison with the known approach rotation invariant shape contexts based on feature-space Fourier transformation proves that the proposed method is faster, more robust to local deformations and more efficient.

Image Registration Using OpponentSIFT Descriptor: Application to Colposcopic Images with Specular Reflections

This paper presents a medical imaging system able to assist practitioner to analyze colposcopic images. The goal is to automatically make all images to a common frame. Image registration has been used to ensure pixel correspondence to the same tissue location throughout the whole temporal sequence. Best approaches are based on using local information, but they are very sensitive to light change and reflections which are frequently current in colposcopic images. In this paper, we propose an approach less sensitive to these variations, moreover, it works well even if reflections are present in colposcopic images. The efficiency and the robustness of the method for colposcopic images are demonstrated.

Robust facial expression recognition system based on hidden Markov models

Facial expressions are considered important communicative tools. In this paper, we present a new system able to recognize facial expressions. This system utilizes one-dimensional Hidden Markov Models (1D-HMMs) with small number of states involving fast computational time during the learning and recognition steps. First, face features are obtained using Gabor wavelets. Then, the Fisher’s Discriminant Analysis method is employed to reduce the features dimensions to eliminate redundancy and overlap in the information. The proposed system differs from previous methods using 1D-HMMs, by the fact that it employs 1D-HMMs with only three states without any prior segmentation step of interest regions in the face (hair, forehead, eyes, etc). The proposed system is evaluated using the JAFFE and KDEF facial expressions data sets.

Image Categorization Using a Heuristic Automatic Clustering Method Based on Hierarchical Clustering

One approach to image categorization is the use of clustering algorithms to sets of images represented by various image descriptors. We propose the use of an automatic clustering algorithm to categorize an image-set represented by color moments. Using this clustering algorithm based on hierarchical clustering, this approach produced adequate results with only minimal user input when applied to a restricted image-set.

A Heuristic Automatic Clustering Method Based on Hierarchical Clustering

We propose a clustering method which produces valid results while automatically determining an optimal number of clusters. The proposed method achieves these results with minimal user input, of which none pertains to a number of clusters. Our methods effectiveness in clustering, including its ability to produce valid results on data sets presenting nested or interlocking shapes, is demonstrated and compared with cluster validity analysis to other methods to which a known optimal number of clusters is provided, and to other automatic clustering methods. Depending on the particularities of the data set used, our method has produced results which are roughly equivalent or better than those of the compared methods.