دانلود رایگان مقاله انگلیسی امنیت با استفاده از پردازش تصویر و شبکه عصبی پیچشی عمیق - IEEE 2018

Abstract

Safety has, for a long time, been one big thing everyone is concerned about. Security breach of private locations has become a threat that everyone intends to eliminate. The traditional security systems trigger alarms when they detect a security breach. However, the usage of image processing coupled with deep learning using convolutional neural networks for image identification and classification helps in identifying a breach in an enhanced fashion thereby increasing security furthermore to a great extent. This is due to its capability to extract complex features from the images using accurate and advanced face and body detection algorithms. The rate at which machine learning, especially deep learning, is transitioning is very high. The use of such technology in taking the existing systems and models to the next level would be a great step towards advancements in every field of science and technology. The same goes with computer vision. These two coupled and brought together to be used in the field of security results in achieving a lot more than what is imagined to be possible and this paper aims to do the same.

VII. CONCLUSION AND FUTURE WORK

Improvement to this system can be done using the OpenFace and the classifier it offers. OpenFace helps us to get the 128 measurements of the face and that is sent as an input to the classifier. Looking at all the measurements of the images which are measured before and the classifier will check with the closest match of the face. This can be further enhanced using the FaceNet model of Google which can produce better results. FaceNet was able to produce an accuracy of about 99.63%. The loss function used to minimize the error is as follows- The above function represents the embedding in a multidimensional space, where x represents the image in the function. The loss is calculated according to the nearest neighbor classifier. This loss function here tries to reduce the distance between the similar images xa i and xp i and away from the other images xn i. Here Į is the margin enforced between the positive and negative images.

The algorithm can be further improved using the blob detection algorithm which aims to detect the areas in images that differ in any property or similar in property. The system can be further enhanced using a new way of memorizing the faces of the people that newly visit the area to be secured which would result in the neural network model to be automatically retrained to adapt to the changes that result during the addition of new images. This also avoids the need for deployment on a server with extremely high computational power since the cost of training after the initial setup is much lesser than initial cost.