Fig 1 - uploaded by E. Yaacoub
Content may be subject to copyright.
Source publication
This paper highlights the work on an Arabic handwriting recognition method, designed with the objective of providing feedback to the user on the correctness of the character written, and in case of incorrect writing, indicating what part of the letter was erroneously written. The target is to combat adult illiteracy in the Arab world by using Infor...
Context in source publication
Context 1
... offline character recognition, after scanning, the word is converted to a binary image or an iconic picture, with each set of pixels representing a certain feature (loop, line, etc.). Features could be further assigned characters. Processing techniques used for analysis can be found, for example, in [8-10]. A detailed overview of how characters are analyzed can be read in [7]. Traditional techniques relied on neural networks (e.g. [3], [11]) for recognition. Other techniques include principal component analysis (PCA) and Fisher linear discriminant (FLD) along with Bayesian classification, as used for example in [12] in order to detect handwritten digits. The objective of the previous techniques is the correct detection of a character from a set of given characters. However, in an educational system, the user, or “learner”, is asked to write a specific letter. Hence, the correct letter is known to the teaching software. Consequently, the recognition software does not need to recognize the typed word or letter from a set of letters or words, but rather indicate to the user if his input was correct or not, and if not, give useful feedback in order for the illiterate user to correct his mistakes. We propose a method based on offline recognition, but that can be easily extended to online. The novelty of the proposed approach resides mainly in this “feedback” characteristic, in addition to its versatility and ease of expandability to the case of general handwriting recognition (i.e. where the software needs to recognize the typed letter from a set of letters and is not aware of the correct result), and to online handwriting recognition. The proposed method is based on the characteristics of each letter which include: loops, bulks, lines, and their relative positions and proportions. The Arabic alphabet with the different forms of each letter is presented in Fig. 1. The details of the proposed approach are detailed in the next ...
Similar publications
Segmenting arabic manuscripts into text-lines and words is an important step to make recognition systems more efficient and accurate. The major problem making this task crucial is the word extraction process: first, words are often a succession of sub-words where the space value between these sub-words do not respect any rules. Second, the presence...
Citations
... For feature extraction geometrical features are extracted which are the basic character image line types and the number of lines appeared in the the zones of the image[2]. So from the set of collected sample letters after cropping and preprocessing is done, the image is divided into windows of equal size, and the feature extraction is done on individual windows. ...
... In this study, the online recognition system is presented , where the writing process is done in a real-time then the character is directly converted into another format as a function of time. When the user writes on screen by using the mouse, the signal is traced into coordinates in function of time: x (t) and y (t) [1]. ...
Inthispaperwepresenta new program (On-lineArabic Language LearningSystem (OALLS)), thatcanbeusedto learn thechildrenhow towritethe arabiccharactersand words,then itexamine iftheycan writethecharacters and thewordsin correctshape. The program isbased on theimage processingtechniquesand itdealswiththe arabiclanguagecharacters.In thissystem we have 40 charactersand 400words,foreach characterand foreach wordwehave4templatesand 25squaredvalues,and the matching prcess based on theintensity ofthe colorsin each field.Thisprogram summarized intwomain parts: alearningpart,and atestingpart.The characterorthe wordwhich iswrittenwillbe matched withitstemplate and show theresultofthetest. This program shows a good resultsand itrecognizecharactersinmany epoches withahighaccuracy(98.60%forCharactersand 96.50% forwords),in which the experiments wereappliedon childrenwithagesbetween6-11years.
The unprecedented growth in applications making use of digital images and multimedia references raised the requirement for image and topic search. Systematic processing of this information is a basic prerequisite for effective analysis of this information, as well as organization and management of it. Likewise, large collections of images are available on the Web, and many search engines provide the possibility of Web image searching based on the keywords. However, there are some problems in finding the desirable image according to the user needs. These problems include inexpressiveness of queries in description of user requirement, large number of irrelevant images to the intended search, lack of summarization, time-consuming review of overall images and lack of diversity. Clustering of image search results can be an efficient solution in solving these problems. In this work, several Folding-based algorithms have been proposed for clustering of image search results. In these algorithms, the efficiency, in comparison with Folding algorithm, is improved through a more effective selection of cluster’s representatives, fuzziness, weight and utilization of hierarchical algorithm preferences. In the represented system, in order to fit clusters with data more appropriately, an algorithm is proposed for refining the clusters. The proposed clustering causes a more convenient task in retrieval process for the user and also causes the efficient retrieval of images. According to the experiences, the proposed method improves the acceptable precision of image clustering.
In this paper an off-line Arabic/Farsi handwritten recognition Algorithm on a subset of Farsi name is proposed. In this system, There is no sub-word segmentation phase. Script database includes 3300 images of 30 Farsi common names. The features are wavelet coefficients extracted from smoothed word image profiles in four standard directions. The Centers of competitive layer of RBF neural network have been determined by combining GA and K_Means clustering algorithm. Weights of supervised layer has been trained by using LMS rule and the distances of feature vector of each sample to the centre of RBF network have been computed based on warping function. Experimental results show advantages of this method in field of handwriting recognition.
