Hassiba Nemmour

Neural network combination by fuzzy integral for robust change detection in remotely sensed imagery

Combining multiple neural networks has been used to improve the decision accuracy in many application fields including pattern recognition and classification. In this paper, we investigate the potential of this approach for land cover change detection. In a first step, we perform many experiments in order to find the optimal individual networks in terms of architecture and training rule. In the second step, different neural network change detectors are combined using a method based on the notion of fuzzy integral. This method combines objective evidences in the form of network outputs, with subjective measures of their performances. Various forms of the fuzzy integral, which are, namely, Choquet integral, Sugeno integral, and two extensions of Sugeno integral with ordered weighted averaging operators, are implemented. Experimental analysis using error matrices and Kappa analysis showed that the fuzzy integral outperforms individual networks and constitutes an appropriate strategy to increase the accuracy of change detection.

Off-line signature verification using artificial immune recognition system

In various pattern recognition applications, artificial immune systems achieve comparable and commonly higher performance than other classification schemes such as SVM. In this paper, we investigate their applicability for handwritten signature verification. Specifically, Ridgelet transform and grid features are used to extract pertinent characteristics. Performance assessment is conducted on the CEDAR dataset comparatively to SVM classifiers. The results in terms of average error rate highlight the high performance of artificial immune recognition algorithm.

New Jaccard-Distance Based Support Vector Machine Kernel for Handwritten Digit Recognition

This paper proposes a new negative Jaccard distance- based kernel for Support Vector Machines (SVM). The Jaccard distance is based on shape comparison between data, which could have a particular importance for handwritten character recognition where each class has its own shape form. So, it seems more proficient than Euclidian distance that is used with conventional kernels. The performance of negative Jaccard kernel is evaluated comparatively to standard SVM kernels for handwritten digit recognition. Experiments are conducted on both One-Against-All (OAA) and One-Against-One (OAO) multi-class SVM implementations using samples taken from USPS database. The results obtained showed that Jaccard Negative Distance kernel outperforms other kernels in most cases.

Kalman filtering as a multilayer perceptron training algorithm for detecting changes in remotely sensed imagery

The multilayer perceptron is usually trained by the backpropagation (BP) algorithm for computing the synaptic weights. In this paper, we investigate the use of Kalman filtering (KF) as a training algorithm for detecting changes in remotely sensed imagery. By using SPOT images and based on some evaluation criteria, the detailed comparison indicates that the KF algorithm is preferable compared to the BP algorithm in terms of convergence rate, stability and change detection accuracy.

Comparison between object and pixel level approaches for change detection in multispectral images by using neural networks

We propose in this paper the investigation of the change detection approaches based on the pixel level and the object level. The pixel level approach is based on the simultaneous analysis of multitemporal data, while the object level approach uses a comparative analysis of independently produced classifications of data. Thereby, the comparison is established by using the multilayer neural network classifier. Usually, the backpropagation algorithm is used as a training rule. In this paper, we investigate the use of the Kalman filtering (KF) as the training algorithm for detecting changes in remotely sensed imagery. By using SPOT images and evaluation criteria, the detailed comparison indicates that the KF algorithm is preferable compared to the BP algorithm in terms of convergence rate, stability and change detection accuracy.

Fuzzy integrals for combining multiple SVM and histogram features for writer's gender prediction

This study addresses automatic prediction of the writers gender. We propose the use of fuzzy integral (FI) operators to combine support vector machines (SVMs) associated with different local features. Presently, we focus on local histogram-based features that describe different kinds of handwriting traits to ensure SVM complementarity. First, we introduce a new feature based on the histogram of templates that aims to highlight local orientations of the text strokes. As a second feature, we propose the rotation invariant uniform local binary patterns to enhance local textural information, whereas the third feature is the gradient local binary patterns. Various forms of the FI are used for combining these predictors. Experiments are conducted on four standard datasets of English, Arabic and French handwritten text. First, for each language, the prediction task is evaluated by considering text-independent and writer-independent design. Then, a more challenging prediction is tried by adding the language-independency constraint. The results obtained confirm the effectiveness of the proposed features. Also, they highlight the contribution of the combination step to achieve a robust prediction.

Fuzzy Integral Based-Mixture to Speed Up the One-Against-All Multiclass SVMS

The one-against-all (OAA) is the most widely used implementation of multiclass SVM. For a K-class problem, it performs K binary SVMs designed to separate a class from all the others. All SVMs are performed over the full database which is, however, a time-consuming task especially for large scale problems. To overcome this limitation, we propose a mixture scheme to speed-up the training of OAA. Thus, each binary problem is divided into a set of sub-problems trained by different SVM modules whose outputs are subsequently combined throughout a gating network. The proposed mixture scheme is based on Sugenos fuzzy integral in which the gater is expressed by fuzzy measures. Experiments were conducted on two benchmark databases which concern handwritten digit recognition (ODR) and face recognition (FR). The results indicate that the proposed scheme allows a significant training and testing time improvement. In addition, it can be easily implemented in parallel

Writer Retrieval Using Histogram Of Templates Features and SVM

In this work, we present a new protocol for a novel biometric scenario that is called writer retrieval. Precisely, we propose to use the Histogram Of Templates (HOT) to generate features from handwritten text images. Then, the retrieval task is achieved by SVM classifier trained according to a writer independent strategy. Experiments are conducted on the CVL database which contains 310 writers (1604 documents written in English and German). The results obtained in terms of overall accuracy highlight the effectiveness of the proposed system.

An-open system for off-line handwritten signature identification and verification using histogram of templates and SVM

Offline signature identification and verification systems encounter several challenges such as the diversity of signatories and the limited number of references. To address these problems we propose a new writer-independent system for signature identification and verification. Besides, a new feature generation scheme is proposed by using the Histogram Of Templates (HOT). The identification and verification step is performed by SVM. Experiments are conducted on a standard dataset which contains off-line signatures of 55 persons. The results obtained are very promising.

Segmentation-verification based on fuzzy integral for connected handwritten digit recognition

This paper investigates a number of verification rules to validate the segmentation of connected handwritten digits. The verification technique based on statistical reasoning and fuzzy integrals is employed to verify the segmentation through decision functions produced by multiclass SVM based recognizers. The segmentation relies on an oriented sliding window which identifies potential cut points. The resulting segmented digits are fed to recognizers and the best segmentation is identified by the verification module that combines the recognizer outputs using fuzzy integrals. The proposed methodology is evaluated on a database of handwritten digits with single as well as multiple connections. Comparative analysis shows that the use of the fuzzy integral allows providing high recognition rates comparatively to the state of the art.