| دانلود رایگان مقاله انگلیسی + خرید ترجمه فارسی
|
|
| عنوان فارسی مقاله: |
ترکیب سریالی شبکه عصبی برای OCR عربی |
| عنوان انگلیسی مقاله: |
A Serial Combination of Neural Network for Arabic OCR |
|
|
| مشخصات مقاله انگلیسی (PDF) | |
| سال انتشار مقاله | 2014 |
| تعداد صفحات مقاله انگلیسی | 7 صفحه با فرمت pdf |
| رشته های مرتبط با این مقاله | مهندسی کامپیوتر |
| گرایش های مرتبط با این مقاله | هوش مصنوعی |
| مجله مربوطه | علوم کامپیوتر (Computer Science) |
| دانشگاه تهیه کننده | آزمایشگاه تحقیقات در سیستم های پیچیده علوم کامپیوتر، دانشگاه Oum El Bouaghi، الجزایر |
| کلمات کلیدی این مقاله | تشخیص عربی، ترکیب سریالی، تابع پایه شعاعی، نظریه رزونانس تطبیقی |
| شناسه شاپا یا ISSN | ISSN 0302-9743 |
| لینک مقاله در سایت مرجع | لینک این مقاله در سایت Springer |
| نشریه اسپرینگر | Springer |
| مشخصات و وضعیت ترجمه فارسی این مقاله (Word) | |
| تعداد صفحات ترجمه تایپ شده با فرمت ورد با قابلیت ویرایش و فونت 14 B Nazanin | 8 صفحه |
| ترجمه عناوین تصاویر و جداول | ترجمه شده است |
| ترجمه متون داخل تصاویر | ترجمه شده است |
| ترجمه متون داخل جداول | ترجمه شده است |
- فهرست مطالب:
چکیده
۱ مقدمه
۲ کار قبلی
۳ ترکیب شبکه های عصبی
۴ نتایج ازمایشی
۵ نتیجه گیری
- بخشی از ترجمه:
طرح کامل تشخیص لغت دست خط عربی نامحدود بر اساس شبکه عصبی ارایه می شود. موتور کلی این ترکیب تابع پایه شعاعی با نظریه رزونانس تطبیقی، سیستمی برای طبقه بندی کلمات دست خط است. این سیستم از عملیات پیش پردازش ، نرمال سازی، تشخیص خطا و تصحیح استفاده کرده و سپس، فاز استخراج ویژگی با ممان تچبیف انجام می شود. سپس، شبکه RBF به صورت اولین موتور تشخیص استفاده می شود که تولید 30 گروه استفاده می شود هر یک دارای 60 تصویر کلمه است. در نهایت، برای هر گروه شبکه ART1 برای طبقه بندی استفاده می شود. سرعت تشخیص حاصله 76.69 درصد است.
نتایج بدست آمده بر اساس جدید بودن ایده علی رغم مسئله یابی در دیتابیس مربوط به نوشتار بد مفید است. مرحله پس پردازش در کار ما بررسی نمی شود و این در تغییرات آینده در سیستم بحث خواهد شد.
- بخشی از مقاله انگلیسی:
Introduction In the last few years many academic institutions and industrial companies have been involved in the field of handwriting recognition. The automatic recognition of handwritten word can be extremely useful in many applications where it is necessary to process large volumes of handwritten data, such as recognition of addresses and postcodes on envelopes, interpretation of amounts on bank checks, document analysis, and verification of signatures. Substantial progress has been recently achieved, but the recognition of handwritten word cannot yet approach human performance. The major difficulties descend from the variability of someone’s calligraphy over time, the similarity of some characters with each other, and the infinite variety of character shapes and writing styles produced by different writers. Furthermore, the possible low quality of the text image, the unavoidable presence of background noise and various kinds of distortions (such as poorly written, degraded, or overlapping characters) can make the recognition process even more difficult. Therefore, handwriting recognition is still an open and interesting area for research and novel ideas. During the 1990’s many methods were proposed for combining multiple classifiers for a single recognition task, with these methods, the focus of the field shifted from the competition among specific statistical, syntactic, or structural approaches to the integration of all these as potential contributing components in a combined system. Many combination methods have been proposed, and the applications to practical problems have proven the advantage of ensemble over individual classifiers [1]. A recent survey [13] categorizes the methods into parallel (horizontal) and sequential (vertical, cascaded) ones. Parallel combination [14] is more often adopted for improving the classification accuracy, whereas sequential combination [10] is mainly used for accelerating the classification of large category set. The proposed classification engine is based on a serial combination of an RBF and an appropriate set of ART1 network. The RBF-based classifier is first used to provide a score for the most likely classes according to feature vector composed of the first 49 Tchebichef moments. Then each ART1 network is applied for each group. Our experimental results done on the IFN/ENIT database provide clear evidence that the proposed combined classifier outperforms either the RBFbased classifier or the ART1-based classifier. The rest of the paper is organized as follows: Section 2 gives previous work concerning serial combination on Arabic handwriting recognition. Section 3 presents an overview of the proposed Multi Classifier System; section 4 is devoted to computational experiments; and section 5 summarises the main conclusions. 2 Previous Work The morphology of the Arabic writing presents some characteristics which are the source of their treatments complexity. The Arabic writing is semi cursive script in its printed and handwritten forms [6]. The characters of an Arabic word (or pseudo-word) are horizontally or vertically ligatured which darken the process of segmentation in characters. The forms of letters change according to their positions in the word. Besides, more than half of the Arabic characters include diacritic points in their shape. In the case of Arabic, the use of the multiple classification schemes is very recent and the number of related publications is not significant. Comparing to the serial combination, there are much interest concerning parallel scheme which is the case in [8], [4], [12], [15] and [16]. We give her related works base on serial combination. In [4], a strategy for Arabic handwritten word recognition has been proposed. The idea is based on a sequential hierarchical cooperation of three classifiers, all of a Markovian type. The first classifier is based on a global description of the word using sequential visual indices. The second classifier is associated with an analytic approach that models the characters deprived of their diacritic dots. The third classifier is associated to the sub-word. In this hierarchical strategy A Serial Combination of Neural Network for Arabic OCR 299 of operation, the rates of recognition of the system exceed 89%. This represents an increase of about 8% with respect to the best performing classifier taken individually.
| دانلود رایگان مقاله انگلیسی + خرید ترجمه فارسی
|
|
| عنوان فارسی مقاله: |
ترکیب سریالی شبکه عصبی برای OCR عربی |
| عنوان انگلیسی مقاله: |
A Serial Combination of Neural Network for Arabic OCR |
|
|
