دانلود مقاله یک سیستم تشخیص چهره بر اساس HRPSM_CNN
چکیده
مقدمه
مطالب مرتبط
روش ارائه شده
آزمایش و بحث
کاربرد در سیستم های بهداشت و درمان
نتیجه گیری
منابع
Abstract
Introduction
Related work
Proposed method
Experiment and discussion
Application in healthcare systems
Conclusion
References
چکیده
تشخیص چهره یک فناوری نوظهور است که کاربردهای مختلفی را در زمینه های مختلف مانند تجزیه و تحلیل تصویر پزشکی، نظارت، شناسایی شخصی و موارد مرتبط با امنیت آشکار می کند. به منظور تشخیص موثر تصاویر از مجموعه داده های شناخته شده، تعدادی الگوریتم تشخیص چهره وجود دارد که در عمل وجود دارند. با این حال، چند مشکل در شناسایی موثر با یک پارامتر راضی مواجه می شوند. اگرچه الگوریتم های مختلفی مانند الگوی باینری محلی (LBP)، کد باینری جهت دار (DBC)، ماشین بردار چند پشتیبانی (Multi-SVM) و شبکه عصبی کانولوشنال (CNN) وجود دارد که برای تشخیص چهره استفاده می شوند، باز هم تشخیص چهره وجود دارد. به خصوص برای پایگاه داده های بزرگ به طور رضایت بخشی به دست نمی آید زیرا تصاویر به دلیل نور ضعیف و همچنین به دلیل انسداد رخ داده در تصاویر راکد تحت تأثیر قرار می گیرند. از این رو، رویکرد جدیدی به نام شبکه عصبی کانولوشنال منطبق با مجموعه نقطه قوی ترکیبی (HRPSM_CNN) برای تشخیص مؤثر چهره‌ها از مجموعه داده‌ها در موقعیت‌های نامحدود پیشنهاد شده است. این روش پیشنهادی در مقایسه با الگوریتم‌های سنتی، ویژگی‌های عملکرد گیرنده بهبود یافته را نشان می‌دهد. این HRPSM_CNN 97 درصد از میزان دقت را برای پایگاه داده ORL و AR و 96 درصد برای پایگاه داده چهره LFW ارائه می دهد که به طور قابل توجهی بالاتر از الگوریتم های سنتی موجود است. الگوریتم پیشنهادی در دستگاه کمکی با اختلال بینایی پیاده‌سازی می‌شود و نتایج نشان‌دهنده تشخیص بهتر در شرایط دشوار مانند نورهای مختلف و شرایط آب و هوایی است.
توجه! این متن ترجمه ماشینی بوده و توسط مترجمین ای ترجمه، ترجمه نشده است.
Abstract
Face recognition is an emerging technology that divulges various applications in diverse fields like medical image analysis, surveillance, personal identification, and security related cases. In order to effectively recognize the images from the known data sets, there are a number of face recognition algorithms which are in practice. However, a few problems are encountered in effective recognition with a satisfied parameter. Even though there are various algorithms like Local Binary pattern(LBP), Directional Binary Code(DBC), Multi Support Vector Machine(Multi- SVM), and Convolutional Neural Network(CNN)which are being used for face recognition, still the face recognition is not achieved satisfactorily especially for the large databases as the images are affected due to poor lighting and also owing to occlusion occurring in the stagnant pictures. Hence, a new approach called Hybrid Robust Point Set Matching Convolutional Neural Network(HRPSM_CNN) is proposed to effectively recognize the faces from the data sets over the unconstrained situations. This proposed method shows enhanced receiver operating characteristics when compared to the traditional algorithms. This HRPSM_CNN provides 97 % of accuracy rate for ORL and AR database and 96 % for LFW face database which are significantly higher than the existing traditional algorithms. The proposed algorithm is implemented in visually impaired assistive device and the results show better recognition under difficult situations like various lighting and weather conditions.
Introduction
In the last decade, the face recognition has evolved enormously since the automatic recognition system became a necessity for various applications like personal identification, security purposes, biometric applications and generating passcodes, etc. In addition to face, various parameters like fingerprint, eyeball, palm, and ear are taken into account for the process which cannot be accessed easily by all [1].Face recognition is generally used for satisfying two things: validation and recalling. It is very difficult to find the differences between the images since almost all the face images seem to be identical. Very few parameters only will change the structure of the image [2]. In this situation, recognizing the face under some uncontrolled situations like poor lighting conditions, motion pictures, and also the position variation becomes a challenging task, even it has been evolved with many enlightened methods. Hence, it grabs the minds of researchers and stimulates curiosity on finding solutions to the challenges of processing the images under occlusion [3].
Conclusion
This paper proposed an efficient method for an effective face recognition under some unmanaged situations. The basic idea of this proposed work is to recognize the image even with the different situations like pose variation, poor lighting and also in moving. In this work, images are pre-processed and then faces are detected by Viola Jones algorithm and finally features are extracted and classified by the proposed algorithm HRPSM_CNN. As the step by step process of images through different layers in hybrid RPSM_CNN, the images are recognized over region of interest. In the pooling layer of RPSM_CNN, key points are extracted and matched which helps in recognizing the partially faced images which also help to classify them. This process in turn reduces the time required for recognition. The proposed algorithm is able to recognize the faces in an efficient manner and finally their parameters are analysed and the same is compared for various data sets like ORL, AR and LFW face data set. With this proposed algorithm of HRPSM_CNN, accuracy rate of 97% which is 1.3% better than the other algorithms is achieved. The data sets ORL, AR and LFW face are analysed with various parameters where all the parameters taken were maintaining or enhancing the face recognition performance in a significant manner. Also the proposed algorithm is applied in visually impaired assistive system, and the results shows better recognition of 95% under difficult situation like various lighting and weather conditions.