Mohamed Ben Halima

Survey on clustering methods: Towards fuzzy clustering for big data

In this report, we propose to give a review of the most used clustering methods in the literature. First, we give an introduction about clustering methods, how they work and their main challenges. Second, we present the clustering methods with some comparisons including mainly the classical partitioning clustering methods like well-known k-means algorithms, Gaussian Mixture Models and their variants, the classical hierarchical clustering methods like the agglomerative algorithm, the fuzzy clustering methods and Big data clustering methods. We present some examples of clustering algorithms comparison. Finally, we present our ideas to build a scalable and noise insensitive clustering system based on fuzzy type-2 clustering methods.

Online Arabic writer identification based on Beta-elliptic model

This paper proposes an automatic text-independent online Arabic writer identification system. The main contribution of our system is to explore the utility of Beta-elliptic model in features extraction for online writer identification, due to the rich output of Beta-elliptic model in terms of graphical, kinematical and biometrical data. The efficiency of the considered features has been evaluated using feed forward neural network classifier. Experimental results on ADAB Database show the performance of the proposed system in online Arabic writer identification task.

NF-SAVO: Neuro-Fuzzy system for Arabic Video OCR

In this paper we propose a robust approach for text extraction and recognition from video clips which is called Neuro-Fuzzy system for Arabic Video OCR. In Arabic video text recognition, a number of noise components provide the text relatively more complicated to separate from the background. Further, the characters can be moving or presented in a diversity of colors, sizes and fonts that are not uniform. Added to this, is the fact that the background is usually moving making text extraction a more intricate process. Video include two kinds of text, scene text and artificial text. Scene text is usually text that becomes part of the scene itself as it is recorded at the time of filming the scene. But artificial text is produced separately and away from the scene and is laid over it at a later stage or during the post processing time. The emergence of artificial text is consequently vigilantly directed. This type of text carries with it important information that helps in video referencing, indexing and retrieval.

MRI brain tumor classification using Support Vector Machines and meta-heuristic method

We present a development of a new approach for automated diagnosis, based on classification of Magnetic Resonance (MR) human brain images. 2D Wavelet Transform and Spatial Gray Level Dependence Matrix (DWT-SGLDM) is used for feature extraction. For feature selection Simulated Annealing (SA) is applied to reduce features size. The next step in our approach is Stratified K-fold Cross Validation to avoid overfitting. To optimize support vector machine (SVM) parameters we use Genetic Algorithm and Support Vector Machine (GA-SVM) model. SVM is applied to construct the classifier. An intelligent classification rate of 95,6522% could be achieved using the support vector machine.

A hybrid method of natural scene text detection using MSERs masks in HSV space color

Text detection in natural scenes holds great importance in the field of research and still remains a challenge and an important task because of size, various fonts, line orientation, different illumination conditions, weak characters and complex backgrounds in image. The contribution of our proposed method is to filtering out complex backgrounds by combining three strategies. These are enhancing the edge candidate detection in HSV space color, then using MSER candidate detection to get different masks applied in HSV space color as well as gray color. After that, we opt for the Stroke Width Transform (SWT) and heuristic filtering. Such strategies are followed so as to maximize the capacity of zones text pixels candidates and distinguish between text boxes and the rest of the image. The non-text components are filtered by classifying the characters candidates based on Support Vector Machines (SVM) using Histogram of Oriented Gradients (HOG) features. Finally we apply boundary box localization after a stage of word grouping where false positives are eliminated by geometrical properties of text blocks. The proposed method has been evaluated on ICDAR 2013 scene text detection competition dataset and the encouraging experiments results demonstrate the robustness of our method.

Cluster Forests Based Fuzzy C-Means for Data Clustering

Cluster forests is a novel approach for ensemble clustering based on the aggregation of partial K-means clustering trees. Cluster forests was inspired from random forests algorithm. Cluster forests gives better results than other popular clustering algorithms on most standard benchmarks. In this paper, we propose an improved version of cluster forests using fuzzy C-means clustering. Results shows that the proposed Fuzzy Cluster Forests system gives better clustering results than cluster forests for eight standard clustering benchmarks from UC Irvine Machine Learning Repository.

Scene text detection images with pyramid image and MSER enhanced

Text detection from images in natural scene is one of the most active research areas. It still remains a challenge for researchers because of the complexity of the image in the wild specifically their background. The state of the text presents also different problems of localization such as size, font, color and orientation. This paper presents a new method based on the location of the concentration areas of text candidates in first step. This step allows a mask is applied, the major objective is to filter the maximum the complex background. We use Otsu technique and edge enhancement by pyramid image in different scales. The second step is to fine detection of candidate characters by maximally stable extremal regions (MSER) based on the luminance that gives more meaning to information merged with enhanced edges and connected to surpass the limits of MSER. Then non-text components are filtered out by the character candidate classification based on DTW using SIFT and HOG features. The false positives are eliminated by geometrical properties of text blocks. Finally we apply boundary box localization after a stage of word grouping. The proposed method has been evaluated on ICDAR 2013 scene text detection competition dataset and the encouraging experiments results can be compared with the latest published algorithms.

An intelligent and robust multi-oriented image scene text detection

Text in image is an important source of information. In this article, we describe an approach for detection of text in the scene file. Our method classify pixels into text and non-text areas using neural network and wavelet transformation. It is divided into two steps: a step offline and online step. The experimental results show the performance of our algorithm.

Cluster forest based fuzzy logic for massive data clustering

This article is focused in developing an improved cluster ensemble method based cluster forests. Cluster forests (CF) is considered as a version of clustering inspired from Random Forests (RF) in the context of clustering for massive data. It aggregates intermediate Fuzzy C-Means (FCM) clustering results via spectral clustering since pseudo-clustering results are presented in the spectral space in order to classify these data sets in the multidimensional data space. One of the main advantages is the use of FCM, which allows building fuzzy membership to all partitions of the datasets due to the fuzzy logic whereas the classical algorithms as K-means permitted to build just hard partitions. In the first place, we ameliorate the CF clustering algorithm with the integration of fuzzy FCM and we compare it with other existing clustering methods. In the second place, we compare K-means and FCM clustering methods with the agglomerative hierarchical clustering (HAC) and other theory presented methods using data benchmarks from UCI repository.

Text detection in natural scene images using two masks filtering

Text detection in natural scenes holds great importance in the field of research and still remains a challenge because of size, various fonts, line orientation, different illumination conditions, weak character and complex background in image. The contribution of the proposed method is filtering out complex backgrounds by utilizing two masks filtering based on text confidence map in the first step and multi-channel maximally stable extremal regions (MSERs) in the second step. Both steps are designed to enhancement, maximize capacity of zones text pixels candidates to distinguish text boxes from the rest of the image. Then non-text components are filtered by the classification of character candidate based on Support Vector Machines (SVM) using HOG features. The false positives are eliminated by geometrical properties of text blocks. Finally we apply boundary box localization after a stage of word grouping. The proposed method has been evaluated on ICDAR 2013 scene text detection competition dataset and the encouraging experiments results demonstrate the robustness of our method.