Efficient Generation of Comprehensive Database for Online Arabic Script Recognition

Abstract

The difficulties in segmenting cursive words into individual characters have shifted the focus of handwriting recognition research from segmentation-based approaches to segmentation-free (holistic) methods. However, maintaining and training large number of prototypes (models) that represent the words in the dictionary make the training process extremely expensive and difficult in computing resources. In this paper we present an efficient system that automatically generates prototypes for each word in a given dictionary using multiple appearance of each letter shape. Multiple appearance allows for many permutation of shapes for each word and thus complicates searching for the right prototype. To simplify the training, reduce the maintained prototypes, and avoid over fitting, we used dimensionality reduction followed by clustering techniques to reduce the size of these sets without affecting their ability to represent the wide variations of the handwriting styles. A set of generated fonts are created by professional writers imitating all handwriting styles for each character in each position. These Fonts are used to generate all shapes for writing each word-part in a comprehensive dictionary. Principal component analysis and k-means clustering techniques are performed to select the minimal number of shapes representing the wide variations of handwriting styles for a word-part. Experimental results using an online recognition system proves the credibility of this process compared to manually generated databases.

Word spotting for handwritten documents using Chamfer Distance and Dynamic Time Warping

“…An additional improvement we have considered is to provide better seeds for starting the clustering process. We plan to use an algorithm for synthetic generation [17] of images of a predefined set of the targeted word-parts, and by doing that, enabling automatic clustering of word-parts, with no need for human operator assistance.…”

Handwriting synthesis: classifications and techniques

“…Although all of their synthesized data were perfectly labeled with its correct script type, the authors commented that the differences between correlation coefficients were quite small and not very reliable. In [12], normal OCR Turing test is used for the evaluation of synthesized Arabic handwriting. The models derived in [48] achieve 99.4 % success rate when tested as recognizers.…”

“…Except for perturbation-based techniques, the two other techniques require shape-matching operations [14,50]. Table 4 presents different works classified by the three generation-based techniques along with the Rao [42] Guyon [20] Jawahar and Balasubramanian [23] Wang et al [1,21,39] Wang et al [1] Lin et al [53] Choi et al [22,47] Saabni [4,12] Jawahar and Balasubramanian [23] Zheng various output data types used. In the following subsections, each of the three generation techniques is discussed in detail.…”

“…In Table 1, we classify the applications of handwriting synthesis by their writer imitation and target aspects. In some cases (e.g., [16]), large databases of handwriting can be synthesized to generate writing samples for a single writer as well as in multi-writer Pen-based PC [1,21] Writer imitation [38,48] Calligraphy arts [27,28] PC personalization [20,22,27,28,44,49] CAPTCHAs [5,43] Machine Text recognition [12,14,15,17,18,28,29,36,38,41,42,48,[50][51][52][53] Writer identification [14,28,46,48] Word spotting [38] Word spotting [38] Compression [9,42] setup. Some researchers have developed systems that can function in either a writer-independent or a writer-specific modes [22,47].…”

“…Among the refereed journals that contribute to the dissemination of established knowledge in the area are the International Journal of Document Analysis and Recognition (IJDAR) (e.g., [2][3][4]), Pattern Recognition (e.g., [5][6][7][8]), Pattern Recognition Letters (e.g., [9]), Machine Learning (e.g., [10]), and others. Besides, some prestigious conferences such as the International Conference on Document Analysis and Recognition (ICDAR) (e.g., [11][12][13][14]), the International Workshop on Document Analysis Systems (DAS) (e.g., [15]), the International Conference on Pattern Recognition (ICPR) (e.g., [16][17][18][19]), and the International Conference on Frontiers in Handwriting Recognition (ICFHR) (e.g., [20][21][22][23]) help in spreading the advances in the field. However, there is a lack of a stand-alone paper that surveys the field.…”