Plant disease detection is one of the most important topics of precision agriculture that aims to protect and boost crop yields. Commonly, intelligent systems based on Convolutional Neural Networks are employed to identify various plant diseases by analyzing images of leaves. For more complex scenarios, such as images captured directly from agricultural fields, advanced deep learning models incorporating attention mechanisms and bottleneck layers are preferred. In this context, the MaxViT model has demonstrated exceptional performance, thanks to its specialized architecture that integrates Mobile Inverted Bottleneck Convolutions along with both block and grid attention mechanisms. Presently, we propose an enhanced version of the MaxViT by incorporating texture features. Specifically, we apply a Gabor filter to the output of the stem convolution block, and the averaged filtered image is then added to the input of the MaxViT block in order to introduce additional texture information. We evaluate our approach on two benchmark datasets of tomato leaf diseases. The first, PlantVillage, contains laboratory-prepared leaf images, while the second, extracted from the PlantWild dataset, includes images captured in wild. Experimental results show that the proposed Gabor-MaxViT model achieves up to a $2.98 \%$ performance improvement over the basic MaxViT and surpasses several state-of-the-art systems.

Accurate and timely diagnosis of plant diseases plays a crucial role in ensuring global food security, safeguarding human health, and supporting the economic well-being of farmers. Currently, there is a growing interest in developing reliable deep learning systems for identifying various plant species and their associated diseases. In this respect, combining multiple CNN is an effective approach for improving disease classification since each individual model has unique design principles that enable them to extract different levels and types of features from input images. The present work introduces a novel fuzzy aggregation-based CNN ensemble for tomato leaf disease classification. Initially, five CNN architectures, namely, VGG16, ResNet152V2, MobileNetV2, InceptionV3, and DenseNet201, are designed for classifying tomato leaf diseases. Based on their diverse statistical characteristics, selected models are then integrated into a CNN ensemble. To achieve this step, we introduce a new fuzzy Max-Sum integral, built upon a fuzzy convex combination operator. Performance assessment is conducted on two datasets including 10 and 6 tomato disease classes. Further experiments are conducted by merging these datasets to include a broader range of disease categories. The results demonstrate that the proposed CNN ensemble outperforms both individual models and the CNN ensemble based on the classical Sugeno fuzzy integral.

Developing robust signature verification systems is one of the most attracting topics in the handwriting recognition field. In fact, dealing with signature spoofing requires the use of robust features that help to distinguish authentic signatures from the forged ones. Presently, we propose a multiscale fusion of two histogram-based features to perform signatures description. Precisely, we introduce the Local Difference Feature as new descriptor that is fused with the Histogram Of Templates. These features are calculated on a multiscale neighborhood to highlight pixels distribution within the signature shape. The verification stage is achieved by using SVM classifier. Performance assessment is carried out on GPDS-300 and MCYT-75 datasets. Results in terms of average error rates evince the robustness of the proposed features, which outperform various state of the art methods.

The present work focuses on combining Convolutional Neural Network (CNN) features to strengthen the writer retrieval in historical document databases. Various CNN models that are LeNet, ResNet, and VGG are used to get writer-independent features of handwritten documents. These features are associated with dissimilarity measures to achieve the retrieval task. Then, developed writer retrieval systems are combined through SVM classifier. Experiments are conducted on ICDAR-2017 dataset which contains historical handwritten documents. The results obtained highlight the robustness of CNN features and the combination stage as well.

A new Histogram Of Templates (HOT) feature describing stroke orientations in handwritten signatures is proposed. Several predefined templates are used to model all possible orientations of handwritten strokes. A comparison between these templates and handwritten strokes is evaluated through a sliding window that considers a central pixel with two neighboring pixels. Presently, a quad-tree partitioning according to the center of gravity is applied to allow a local calculation of HOT. The verification task is achieved by SVM classifier. Experiments are conducted on the well-known CEDAR and MCYT-75 databases. The comparison with the state-of-the-art highlights the effectiveness of the proposed feature.

One-Class Classifier (OCC) has been widely used for its ability to learn without counterexamples. Its extension for multi-class implementation offers an open scheme which allows easily adding new classes. However, using OCCs for the multi-class implementation usually achieves less accuracy than the usual multi-class implementations. Hence, in order to improve the accuracy and keep the offered advantage, the suitable approach consists to combine different classifiers. Thus, this paper proposes a study of combining different types of OCC for multi-class classification by means of a new Dynamic Weighted Average (DWA) combination rule. Experimental results conducted on several real-world datasets prove the effective use of the proposed approach where the DWA rule achieves the best results against fixed rules as well as the decision template. Furthermore, comparison of the proposed open classification system against a standard open classifier based on K-Nearest Neighbor (K-NN) shows the superiority of the proposed system.

This work addresses soft-biometrics prediction from handwriting analysis, which aims to predict the writer's gender, age range and handedness. Three SVM predictors associated each to a specific data feature are developed and subsequently combined to aggregate a robust prediction. For the combination step, Sugeno's Fuzzy Integral is proposed. Experiments are conducted on public Arabic and English handwriting datasets. The performance assessment is carried out comparatively to individual systems as well as to max and average rules, using independent and blended corpuses. The results obtained demon-strated the usefulness of the Fuzzy Integral, which provides a gain of more than 4% over individual systems as well as other combination rules. Moreover, with respect to the state of the art methods, the proposed approach seems to be much more relevant.

This study demonstrates how the combination of oriented Basic Image Features (oBIFs) with the background concavity features can be effectively employed to enhance the performance of isolated digit recognition systems. The features are extracted without any size normalization from the complete image as well as from different regions of the image by applying a uniform grid sampling to the image. Classification is carried out using one-against-all support vector machine (SVM) while the experimental study is conducted on the standard CVL single digit database. A series of evaluations using different feature configurations and combinations realized high recognition rates which are compared with the state-of-the-art methods on this subject.

Usual handwritten signature verification systems address the writer-independent (WI) approach using only bi-class robust classifiers to deal with the most challenging tasks. Indeed, WI concept, the reduced size of references and one-class signature verification are still open issues in practical cases. In this paper, we propose a one-class WI system using feature dissimilarity measures thresholding for classification and a reduced number of references. The proposed system involves the use of contourlet transform-based directional code co-occurrence matrix feature generation method. The verification is achieved through a WI threshold which is automatically selected using a new signature stability criterion. The proposed WI concept is besides addressed through the mixture of different writers' data sets in both the design and verification stages. Experimental results show the effectiveness of the proposed system in spite of the strict verification protocol using the one-class concept, a unique threshold for accepting or rejecting a questioned signature, the reduced number of writers, and the limited number of reference signatures.

This work introduces two gradient features for writer's gender, handedness, and age range prediction. The first feature is the Histogram of Oriented Gradients, which highlights the distribution of gradient orientations within images. The second feature is the so-called gradient local binary patterns, which is an improved gradient feature that incorporates the local binary pattern neighborhood in the gradient calculation. The prediction task is achieved by using SVM classifier. Experiments are performed on two corpuses of English and Arabic handwritten text. The results obtained in terms of classification accuracy highlight the effectiveness of the proposed features, which overcome the state of the art.

Since 2009, DIBCO is becoming the main benchmarking for evaluating objectively the performance of binarization methods by means of various quantitative measures. The usual evaluation protocol is performed on blind datasets, which contain degradations with unknown occurrence and variety. This leads to generate every year random ranking depending on the used dataset. This paper aims to propose a comparative analysis for evaluating binarization methods on unblind dataset based on a new degradation typology and takes into account the written types. Unblind dataset is produced by categorizing DIBCO images according to the dominant degradation and written (handwritten or printed) types. Experiments conducted on blind and unblind DIBCO images show that a binarization method can deal better than others when it is performed on a specific degradation or written type.

This paper introduces an improved implementation of Artificial Immune Recognition System (AIRS) to solve the automatic off-line handwritten signature verification. Conventionally, the AIRS training provide a set of memory cells that are used with a k-Nearest Neighbors decision to classify test patterns. In order to improve the verification ability, we propose to substitute the k-NN classification by a trainable decision function using SVM classifier. In addition, for signature characterization, new gradient local binary pattern features are introduced. Experiments are conducted on CEDAR and GPDS-300 corpuses. The results show that the proposed algorithm overcomes the conventional AIRS-kNN by more than 9% in the average error rate. Also, it gives similar and sometimes better performance than the state of the art.

This work focuses on automatic prediction of the writer's biometrics including gender, handedness and age information. The proposed prediction system is based on the use of Histogram of Oriented Gradients (HOG), which aims to extract gradient directions from the handwritten text. The prediction task is achieved using SVM classifier. Experiments performed on IAM and KHATT datasets, reveal promising results. Precisely, the comparison with the state of the art shows that, the proposed prediction system provides quite encouraging performance.

This work focuses on automatic off-line handwritten signature verification where a new gradient feature is proposed for signature characterization. This feature namely, Gradient Local Binary Patterns (GLBP) takes advantage from textural information, to improve the gradient description within images. The verification step is performed in a writer-dependent framework using SVM classifier. Experimental analysis is conducted on CEDAR and GPDS-300 datasets. The results obtained in terms of average error rate highlight the high performance of the proposed feature, which significantly overcomes several state of the art results.

We propose in this paper a system to recognize handwritten digit strings, which constitutes a difficult task because of overlapping and/or joining of adjacent digits. To resolve this problem, we use a segmentation-verification of handwritten connected digits based conjointly on the oriented sliding window and support vector machine (SVM) classifiers. The proposed approach allows separating adjacent digits according the connection configuration by finding at the same time the interconnection points between adjacent digits and the cutting path. SVM-based segmentation-verification using the global decision module allows the rejection or acceptance of the processed image. Experimental results conducted on a large synthetic database of handwritten digits show the effective use of the oriented sliding window for segmentation-verification.

This paper introduces a new offline handwriting database that was developed to be employed in performance evaluation, result comparison and development of new methods related to handwriting analysis and recognition. The database can particularly be used for signature verification, writer recognition and writer demographics classification. In addition, the database also supports isolated digit recognition, digit/text segmentation and recognition and similar related tasks. The database comprises 600 Arabic and 600 French text samples, 1300 signatures and 21,000 digits. 100 Algerian individuals coming from different age groups and educational backgrounds contributed to the development of database by providing a total of 1300 forms. The database is also accompanied with ground truth data supporting the evaluation of the aforementioned tasks. The main contribution of the database is providing a multi-script platform where same authors contributed samples in French and Arabic. It would be interesting to explore applications like writer recognition and writer demographics classification in a multi-script environment.

This paper investigates the combination of different statistical and structural features for recognition of isolated handwritten digits, a classical pattern recognition problem. The objective of this study is to improve the recognition rates by combining different representations of non-normalized handwritten digits. These features include some global statistics, moments, profile and projection based features and features computed from the contour and skeleton of the digits. Some of these features are extracted from the complete image of digit while others are extracted from different regions of the image by first applying a uniform grid sampling to the image. Classification is carried out using one-against-all SVM. The experiments conducted on the CVL Single Digit Database realized high recognition rates which are comparable to state-of-the-art methods on this subject.

Most of the classical methods for degraded document binarization are based on the pixel gray level intensity or on simple pixel neighborhood information such as mean or variance to compute the binarization threshold. Moreover, these information are extracted from the spatial domain of the document image which are not very discriminative. In this paper, we propose to estimate texture information based on Gabor filters for ancient degraded documents. First, the dominant slant angle of the document image script is computed by using the Fourier transform. Then, this dominant angle is used within a weighted sum of angles in a Gabor filter bank in order to capture more efficiently the document image foreground (text). This information, combined with the variance and the mean extracted respectively from spatial and frequency domains are used for estimating the binarization threshold. Three variants are used for evaluating the performance of Gabor filter bank, which are based on Niblack's, Sauvola's, and Wolf's thresholds. Experimental results conducted on DIBCO Datasets show that the proposed method is more appropriate for poor contrasted documents and ink-bleed through degradations.

The One-Class Classifier (OCC) has been widely used for solving the one-class and multi-class classification problems. Its main advantage for multi-class is offering an open system and therefore allows easily extending new classes without retraining OCCs. However, extending the OCC to the multi-class classification achieves less accuracy comparatively to other multi-class classifiers. Hence, in order to improve the accuracy and keep the offered advantage we propose in this paper a Multiple Classifier System (MCS), which is composed of different types of OCC. Usually, the combination is performed using fixed or trained rules. Generally, the static weighted average is considered as straightforward combination rule. In this paper we propose a dynamic weighted average rule that calculates the appropriate weights for each test sample. Experimental results conducted on several real-world datasets proves the effective use of the proposed multiple classifier system where the dynamic weighted average rule achieves the best results for most datasets versus the mean, max, product and the static weighted average rules.

The limited number of writers and the lack of forgeries as counterexample to construct the systems is the main difficulty task for designing a robust off-line Handwritten Signature Verification System (HSVS). In this paper, we propose to study the influence of writer's number using conjointly the curvelet transform and the One-Class Support Vector Machine (OC-SVM), which takes in consideration only genuine signatures. The design of the HSVS is based on the writer-independent approach. Experimental results conducted on the standard CEDAR and GPDS datasets demonstrate that the proposed method allows achieving the lowest Average Error Rate with a limited number of writers.

One-Class Classification (OCC) based on the Auto-Associative Neural Networks (AANN) has been widely used in various recognition applications for its effective robustness. Its main advantage lies in the description of samples more accurately to other OCCs. However, it is considerably sensitive to the presence of outliers or noisy data contained into the training set, which may affect badly the representative model. Hence, we propose in this paper an algorithm that uses the AANN for selecting the most representative training samples. The same AANN is retrained to reproduce the selected samples for generating an optimal representative model. The experimental evaluation conducted on several real-world benchmarks confirms the effective use of the Selected Training Samples for Associative Neural Network (STS-AANN) versus the training on the entire set.

We propose in this work a new handwritten digit recognition system based on parallel combination of SVM classifiers for managing conflict provided between their outputs. Firstly, we evaluate different methods of generating features to train the SVM classifiers that operate independently of each other. To improve the performance of the system, the outputs of SVM classifiers are combined through the Dezert-Smarandache theory. The proposed framework allows combining the calibrated SVM outputs issued from a sigmoid transformation and uses an estimation technique based on a supervised model to compute the belief assignments. Decision making is performed by maximizing the new Dezert-Smarandache probability. The performance evaluation of the proposed system is conducted on the well known US Postal Service database. Experimental results show that the proposed combination framework improves the recognition rate even when individual SVM classifiers provide conflicting outputs.

In this paper, we propose a system to recognize handwritten digit strings, which constitutes a difficult task because of overlapping and/or joining of adjacent digits. To resolve this problem, we use a segmentation-recognition of handwritten connected digits based on the oriented sliding window. The proposed approach allows separating adjacent digits according the connection configuration by finding at the same time the interconnection points between adjacent digits and the cutting path. The segmentation-recognition using the global decision module allows the rejection or acceptance of the processed image. Experimental results conducted on the handwritten digit database NIST SD19 show the effective use of the sliding window for segmentation-recognition.

We address, in this work, a new feature generation method for two different approaches of off-line handwritten signature verification (HSV), writer-dependent and writer-independent HSV. The proposed method uses conjointly the contourlet transform and the co-occurence matrix. The contourlet transform allows capturing contour segment directions of the handwritten signature, while the co-occurrence matrix allows describing the number of directions. Experiments are conducted on the well known CEDAR dataset and the classification through the support vector machines (SVM). The obtained results show the effective use of the Contourlet transform for handwritten signature verification comparatively to the state of the art.

A combination handwritten signature verification system is proposed for managing conflicts provided from each individual off-line and on-line support vector machine (SVM), respectively. Basically, the system is divided into three parts: (i) Off-line verification system, (ii) on-line verification system and (iii) combination module using belief function theory. The proposed framework allows combining the normalized SVM outputs and uses an estimation technique based on the dissonant model of Appriou to compute the belief assignments. Combination is performed using belief models such as Dempster-Shafer (DS) rule and proportional conflict redistribution (PCR) rule followed by the likelihood ratio-based decision making. Experiments are conducted on the well-known NISDCC signature collection using false rejection and false acceptance criteria. The obtained results show that the proposed combination framework using Dezert-Smarandache (DSm) theory yields the best verification accuracy even when individual off-line and on-line classifications provide conflicting results.

The objective of this work is to present a signature verification system based on combination of off-line and online systems for managing conflict provided by the Support Vector Machine (SVM) classifiers. This system is basically divided into three parts: i) off-line verification stage, ii) on-line verification stage and iii) combination module using Dempster-Shafer theory (DST). The proposed framework allows combining the normalized SVM outputs and uses an estimation technique based on the dissonant model of Appriou to compute the belief assignments. Combination is performed using Dempster-Shafer (DS) rule followed by the likelihood ratio based decision making. Experiments are conducted on the well know NISDCC signature collection using false rejection and false acceptance criteria. The obtained results show that the proposed combination framework using DST yields the best verification accuracy compared to the sum rule even when individual off-line and on-line classifications provide conflicting results.

In the present work, we investigate the combination of Ridgelet transform and tangent similarities for handwritten Arabic word recognition. Ridgelets are used for generating pertinent features of handwritten words. These features are handled through tangent similarities to enforce the discrimination power by using a priori knowledge. The One-Against-All SVMs implementation is used for the classification stage. Experiments are conducted on a vocabulary of twenty-four words extracted from the IFN/ENIT database. In a first step, the Ridgelet performance is assessed comparatively to the results obtained for uniform grid (zoning) features. Thenafter, tangent similarities are computed to put the variability knowledge into Ridgelet features. Results showed that The combination of tangent similarities and Ridgelet features yields a robust descriptor, which improves the recognition accuracy while accelerating the runtime.

Discrimination of machine printed and handwritten text is deemed as major problem in the recognition of the mixed texts. In this paper, we address the problem of identifying each type by using the Radon transform and Support Vector Machines, which is conducted at three steps: preprocessing, feature generation and classification. New set of features is generated from each word using the Radon transform. Classification is used to distinguish printed text from handwritten. The proposed system is tested on IAM databases. The recognition rate of the proposed method is calculated to be over 98%.

This paper proposes the use of tangent vectors for improving handwritten digit recognition accuracy. Basically, tangent vectors and calculated from the training set are combined with mean vectors of numeral classes in order to generate synthetic training data. The performance of the enlarged database is compared to that of the original one using samples of USPS database. An artificial neural network is used to perform the recognition task. The results showed that the use of tangent vectors data, improves the accuracy to more than 2%.

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 Sugeno's 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

In the last recent years, there was an important interest in using mixture of different support vector machines (SVMs) for large scale data bases. This approach aims to reduce the complexity of the training algorithm of SVMs which is at least quadratic to the number of training data. Commonly, this objective is reached by dividing the data set into small subsets which are much easy to learn. In this paper, we present a new approach for mixture of SVMs based on the notion of fuzzy integral. In this method, a fuzzy measure is used as a gater to avoid the time complexity caused by conventional gaters such as neural networks. Experiments were conducted on USPS handwritten digits data base whose learning is a challenging task. The results obtained indicate that the proposed scheme improves significantly the training time while keeping accuracy at least as good as the accuracy of a single SVM.

We propose in this paper a method for combining simultaneously multispectral (XS), panchromatic (P) and radar (SAR) images by suing conjointly the Intensity Hue Saturation (HIS) transform and the Redundant Wavelet Decomposition (RDW). The fusion is accomplished through two steps. Firstly, XS images are represented in the HIS system in order to isolate the spectral components (H,S) from the spatial information (I) component. Secondly, features coming from P and SAR images are incorporated into the (I) component by using the RWD. The obtained fused images show the contribution of the P image for discriminating lines and edges. While the SAR image provides an information on the surface roughness and the detection of point targets as houses, boats, and water-land transitions.

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.