Online Arabic handwriting recognition using continuous Gaussian mixture HMMS
Citations
94 citations
Cites background or methods from "Online Arabic handwriting recogniti..."
...Combining online and offline preprocessing is also used in [46,47] and [48]....
[...]
...Whereas Al-Habian and Assaleh [47] present a recognizer structure aimed at recognizing online Arabic handwriting written in continuous form, the basic units of recognition used are strokes, which are subletter parts....
[...]
...Offline features were inspected by Al-Habian and Assaleh [47]....
[...]
...Elanwar et al. [51] Geometric features based on Feeman chain + segmentation-based approach using dynamic programming and template matching 317 words (1,814 characters), written by four writers for training 74 % word based 94 words (435 characters) written by other four writers for test 95.4 % character based Kherallah et al. [42] Modeling based on inflection point detection, the overlapped form of beta signals, and the elliptic arcs + beta-elliptical modeling + combining MLPNN + SOM + FKNN 30,000 Arabic digits 95.08 % Al-Taani and Hammad [46] Identifying the changes in the slope’s signs around zero + template matching 3,000 Arabic digits written by 100 persons 95 % Izadi et al. [38] Wavelet-based smoothing technique + Segmentationbased approach + DTW classifier 20 classes of paws with two and three characters for Persian script 89.4 % for two letters word 85 % for three letters word Mezghani et al. [35] + Bayes classification Zhu et al. tangents and histograms projection 528 characters of each letter from each of 22 writers for a total of 9,504 characters 92.61 % No diacritical points The training set contains 6,336 samples and the testing set 3,168 samples Sternby et al. [6] Template matching + using BLSTM algorithm, for dynamically treating the diacritical marks 1,578 samples of 66 Arabic words written by 40 persons Between 80 and 91 % Assaleh et al. [70] The motion information of the hand movement is projected onto two static AD images + video-based approach + KNN classifier Videos of 28 isolated Arabic letters Each letter was written eight times by two different users 97.77 % with polar ADs and 99.11 % for the two-tier-weighted AD scheme Daifallah et al. [37] Segmentation approach + HMM letters without marks or points 150 words 720 letters inside words 85.3–92.6 % for words 88.8– 97.2 % for letters Kherallah et al. [57] Combining visual coding and genetic algorithm 500 words written by 24 persons 97 % for isolated arabic words Saabni and El-Sana [44] Holistic approach + dynamic time warping classification....
[...]
...neural network [36,56,61], k-nearest neighbor [37,42], and other combination of techniques [47,62,75]....
[...]
76 citations
55 citations
Cites methods from "Online Arabic handwriting recogniti..."
...Based on the HMMs system, Al-Habian and Assaleh [17] discuss the use of feature vector extracted from a sliding window on the reconstituted image of the on-line text....
[...]
27 citations
Cites background or methods from "Online Arabic handwriting recogniti..."
...In Alijla and Kwaik [2012], they used the density, the aspect ratio, and character alignment ratio....
[...]
...In Alimi [1997], the segmentation is performed manually during the training phase....
[...]
...Recognition based on sub-character graphemes reduces the number of basic classes, as in Al-Habian and Assaleh [2007]. However, some effort is required to produce meaningful text from these graphemes....
[...]
...In Alijla and Kwaik [2012], they used the density, the aspect ratio, and character alignment ratio. In Ramzi and Zahary [2014], the extraction of offline features is accomplished over three steps: zoning, traversal, and determining the types of line segments. The feature vector is formed by combining the features of the zones. The features of each zone are the normalized length of each line type, the normalized area, Euler number, regional area, and eccentricity. A summary of the features used in some studies is given in Table IV. As shown in Table IV, different types of features are typically combined due to the shortcomings of each type individually. In Ramzi and Zahary [2014], several experiments with different types of features are conducted to select the best features....
[...]
...Post-processing is conducted in two steps in Al-Habian and Assaleh [2007]. First, rules are imposed on the letter shapes to exclude the recognized candidate letter shapes that are invalid for a corresponding position in a word....
[...]
24 citations
Cites background or methods from "Online Arabic handwriting recogniti..."
...Also, AI-habian and Assaleh [5] presented a structured model for recognizing online Arabic handwriting written in continuous form based on Hidden Markov Models (HMMs) to recognize Arabic strokes....
[...]
...[5] G. AI-Habian, and K. Assaleh, "Online Arabic handwriting recognition using continuous Gaussian mixture HMMS, " International Conference on Intelligent and Advanced Systems, ICIAS 2007, Kuala Lumpur, Malaysia. vol: 1, page(s): 1183-1186....
[...]
...According to Al-habian and Assaleh [5], after acquiring the text via input device, a formatting sequence (x, y) was done to represent the text....
[...]
References
21,819 citations
2,653 citations
"Online Arabic handwriting recogniti..." refers background in this paper
...The difference between the two modes is that the online mode provides us with temporal features that are used to infer the dynamics of the writing....
[...]
503 citations
167 citations
"Online Arabic handwriting recogniti..." refers background in this paper
...However, in online recognition the input is being captured sample by sample from the writer while he/she is writing....
[...]
153 citations
"Online Arabic handwriting recogniti..." refers background in this paper
...However, in online recognition the input is being captured sample by sample from the writer while he/she is writing....
[...]