Mounir Ait Kerroum

Textural feature selection by joint mutual information based on Gaussian mixture model for multispectral image classification

Textural features play increasingly an important role in remotely sensed images classification and many pattern recognition applications. However, the selection of informative ones with highly discriminatory ability to improve the classification accuracy is still one of the well-known problems in remote sensing. In this paper, we propose a new method based on the Gaussian mixture model (GMM) in calculating Shannons mutual information between multiple features and the output class labels. We apply this, in a real context, to a textural feature selection algorithm for multispectral image classification so as to produce digital thematic maps for cartography exploitation. The input candidate features are extracted from an HRV-XS SPOT image of a forest area in Rabat, Morocco, using wavelet packet transform (WPT) and the gray level cooccurrence matrix (GLCM). The retained classifier is the support vectors machine (SVM). The results show that the selected textural features, using our proposed method, make the largest contribution to improve the classification accuracy than the selected ones by mutual information between individual variables. The use of spectral information only leads to poor performances.

Textural feature selection by mutual information for multispectral image classification

Selecting the most informative features from high dimensional space is one of the well-known problems in multispectral image classification and pattern recognition applications. The commonly used techniques for dimensionality reduction are the Principal Components Analysis (PCA) and the Linear Discriminant Analysis (LDA). However, their components are not necessarily the best for such classification. In this work, we investigate the effectiveness of two Mutual Information Feature Selector (MIFS) algorithms. The objective is to identify the most optimal algorithm that reduce the important dimension of input textural feature space while keeping the highest accuracy classification. The candidate textural features are extract, from an HRV-XS SPOT image of forest region in Rabat, Morocco, using Wavelet Transform (WT) at level (l=1,2). Experimental results prove that MIFS algorithms give a better performances, in terms of dimensionality reduction and classification accuracy, than classical methods PCA and LDA. The retained classifier is the Support Vectors Machine (SVM).

Combining classifiers using Dempster-Shafer evidence theory to improve remote sensing images classification

Classification system and textural features play increasingly an important role in remotely sensed images classification and many pattern recognition applications. In this work, we propose to fuse the information outputed by 3 well-known classifiers : Support Vector Machines (SVM), Neural Network (NN) and Parzen Window (PW). These classifiers were combined according to the Dempster-Shafer theory. The input textural feature used are selected according the GMMFS algorithm [1]. The classification results show that the proposed method gives high performances than those of classifiers separately considered.

Using the maximum Mutual Information criterion to textural Feature Selection for satellite image classification

This paper presents and evaluates the use of the maximum mutual information criterion to textural feature selection for satellite image classification. Our approach is based on a recent work of Mutual Information Feature Selector Algorithm. The effectiveness of the proposed approach is evaluated on real data. In fact, the textural features are extracted using the cooccurrence matrix from two forest zones of SPOT HRV(XS) image in the region of Rabat, Morocco. The experimental tests of this study prove that the proposed approach gives a better performance for satellite image classification than classical methods such as principal components analysis (PCA) and linear discriminant analysis (LDA). The classifier used in this work is the support vectors machine (SVM).

Variational image segmentation models: Application to medical images MRI

Image segmentation is an important branch of computer vision. Its aim is to extract meaningful lying in objects images, either by dividing images into contiguous semantic regions, or by extracting one or several objects more specific in images, such as medical structures. In general, image segmentation task is very difficult to achieve it since natural images are diverse, complex and the way we perceive them, vary according to individuals. More than a decade ago, a promising mathematical framework, based on variational models and partial differential equations, have been investigated to solve the image segmentation problem. This new approach benefits from well-established mathematical theories that allow people to analyze, understand and extend segmentation methods. Moreover, this framework is defined in a continuous setting which makes the proposed models independent with respect to the grid of digital images.

Texture classification based on curvelet transform and extreme learning machine with reduced feature set

In this work, a novel approach for texture classification is proposed. We present a highly discriminative and simple descriptor to achieve feature learning and classification simultaneously for texture classification. The proposed method introduces the application of digital curvelet transform and explores feature reduction properties of locality sensitive discriminant analysis (LSDA) in conjunction with extreme learning machine (ELM) classifier. The image is mapped to the curvelet space. However, the curse of dimensionality problem arises when using the curvelet coefficients directly and therefore a reduction method is required. LSDA is used to reduce the data dimensionality to generate relevant features. These reduced features are used as the input to ELM classifier to analytically learn an optimal model. In contrast to traditional methods, the proposed method learns the features by the network itself and can be applied to more general applications. Extensive experiments conducted in two different domains using two benchmark databases, illustrate the effectiveness of the proposed method. In addition, empirical comparisons of the proposed method against curvelet transform in conjunction with traditional dimensionality reduction tools show that the suggested method does not only lead to a more reduced feature set, but it also outperforms all the compared methods in terms of accuracy.

Multi-stage fusion of local and global features based classification for traffic sign recognition

The automatic traffic sign detection and recognition (TSDR) provide an additional level of driver assistance, leading to increase passengers, road users and vehicles safety. As part of Advanced Driving Assistance Systems (ADAS), traffic sign recognition (TSR) has drawn considerable research attention in recent years due to its challenging nature as a computer vision problem. It is usually tackled in three stages: detection, feature extraction and classification. This paper focuses on the second stage of the process, namely traffic sign feature extraction and proposes to fuse two discriminative and complementary feature sets. In this approach, Discrete Cosine Transform (DCT) is used to extract global features of traffic sign while Local Binary Patterns (LBP) is used to extract local descriptors. The classification of these features is performed using the Support Vector Machine (SVM). The proposed fusion approach is validated on the German Traffic Sign Recognition Benchmark Dataset (GTSRD) and has been found to be more efficient than a recognition system which uses only one feature, trained individually.

An improved fuzzy clustering approach using possibilist c-means algorithm: Application to medical image MRI

Currently, the MRI brain image processing is a vast area of research, several methods and approaches have been used to segment these images (thresholding, region, contour, clustering). In this work, we propose a novel segmentation approach, which is based on fuzzy c-means clustering and using possibilist c-means approach. To validate our approach, we have tested successfully on several datasets of real images MRI. Thus, to show the performance of our method, we compared our results with different segmentation algorithms: k-means, fuzzy c-means, and possibilist c-means.

Band selection by mutual information for hyper-spectral image classification

Band selection is one of the most important problems in hyper-spectral image classification. Indeed, the presence of irrelevant and/or redundant bands can harm the performance of classification accuracy. This paper investigates the effectiveness of four mutual information feature selector MIFS algorithms to select the informative bands for hyper-spectral image classification. These algorithms are: MIFS, MIFS-U, MIFS-U2 and NMIFS. Our main motivation behind the study of this family algorithm is due to the fact that mutual information MI is an effective indicator to measure the overall statistical dependency between variables and it has proved its efficiency in many pattern recognition problems, especially in remote sensing. The experimental results have been made on two AVIRIS hyper-spectral datasets Indian Pines and Salinas and prove that MIFS algorithm and its variants give promising performances, in terms of dimensionality reduction and classification accuracy than MI-est method Guo et al., 2006, specially for high dimensional data with many irrelevant and/or redundant bands.

Word-based Arabic handwritten recognition using SVM classifier with a reject option

Arabic handwritten recognition is a challenging task due to high variability of Arabic script and its intrinsic characteristics such as cursiveness, ligatures and diacritics. This paper presents a word-based off-line Arabic handwritten recognition system based on discrete cosine transform features and SVM classifier enhanced using a reject option. The latter is based on the number of sub-words in the input word image calculated using a novel segmentation algorithm. To evaluate our proposed system, we used the IFN/ENIT database of Arabic handwritten words and the results has shown the effectiveness of our approach in enhancing the recognition performance.