Tomoka Azakami

Bio: Tomoka Azakami is an academic researcher from Tokyo University of Technology. The author has contributed to research in topics: CAPTCHA & Convolutional neural network. The author has an hindex of 2, co-authored 5 publications receiving 9 citations.

Challenge of Deep Learning against CAPTCHA with Amodal Completion and Aftereffects by Colors

Abstract: We make experiments of machine learning for our CAPTCHA proposed as an effective CAPTCHA with amodal completion and aftereffects by colors. CAPTCHA is a method that distinguishes human beings from artificial intelligence in order to prohibit malicious programs from acquiring accounts on Internet. The most popular CAPTCHA is text-based CAPTCHA with distorted alphabets and numbers. However, it is known that all of text-based CAPTCHA algorithms can be analyzed by computers. In addition, too much distortion or noise prevents human beings from recognizing alphabets or numbers. As a solution of the problems, an effective text-based CAPTCHA algorithm with amodal completion was proposed by our team. Our CAPTCHA causes computers a large amount of calculation costs while amodal completion helps human beings to recognize characters momentarily. Our CAPTCHA has evolved with aftereffects and combinations of complementary colors. In this paper, we evaluate our CAPTCHA by machine learning since machine learning is faster and more accurate than existing calculations by a computer. We confirm the limitations of machine learning. Especially, we focus on whether a computer can recognize characters without knowledge of amodal completion.

Effective CAPTCHA with Amodal Completion and Aftereffects by Complementary Colors and Difference of Luminance

Abstract: We propose an effective CAPTCHA with amodalcompletion and aftereffects and improve it by applying complementary colors and difference of luminance. CAPTCHA is a method which distinguishes human beings from artificial intelligence which is usually called bots on Internet. Bots acquire accounts of web services for unfair uses such as stealth marketing or SPAM attacks. That is, CAPTCHA is a countermeasure against those attacks. The most popular kind of CAPTCHA methods is text-based CAPTCHA in which distorted alphabets and numbers appear with obstacles or noise. However, it is known that all of text-based CAPTCHA algorithms can be analyzed by computers. In addition, too much distortion or noise prevents human beings from alphabets or numbers. There are other kinds of CAPTCHA methods such as image CAPTCHA and audio CAPTCHA. However, they also have problems in use. As a related work, an effective text-based CAPTCHA algorithm with amodal completion was proposed. The CAPTCHA provides computers a large amount of calculation cost while amodal completion helps human beings to recognize characters momentarily. On the other hand, extreme concentration is required for momentary recognition. The CAPTCHA is once improved in our laboratory. Aftereffects are additionally combined with the CAPTCHA algorithm. Moreover, in this paper, we add colors to the algorithm with considering combinations of complementary colors and those of luminance. The aftereffects extend time for recognition of characters from a moment to several seconds.

Creation of Adversarial Examples with Keeping High Visual Performance

Abstract: The accuracy of the image classification by the convolutional neural network is exceeding the ability of human being and contributes to various fields. However, the improvement of the image recognition technology gives a great blow to security system with an image such as CAPTCHA. In particular, since the character string CAPTCHA has already added distortion and noise in order not to be read by the computer, it becomes a problem that the human readability is lowered. Adversarial examples is a technique to produce an image letting an image classification by the machine learning be wrong intentionally. The best feature of this technique is that when human beings compare the original image with the adversarial examples, they cannot understand the difference on appearance. However, Adversarial examples that is created with conventional FGSM cannot completely misclassify strong nonlinear networks like CNN. Osadchy et al. have researched to apply this adversarial examples to CAPTCHA and attempted to let CNN misclassify them. However, they could not let CNN misclassify character images. In this research, we propose a method to apply FGSM to the character string CAPTCHAs and to let CNN misclassified them.

Challenge to Impede Deep Learning against CAPTCHA with Ergonomic Design

Abstract: Once we had tried to propose an unbreakable CAPTCHA and we reached a result that limitation of time is effect to prevent computers from recognizing characters accurately while computers can finally recognize all text-based CAPTCHA in unlimited time. One of the existing usual ways to prevent computers from recognizing characters is distortion, and adding noise is also effective for the prevention. However, these kinds of prevention also make recognition of characters by human beings difficult. As a solution of the problems, an effective text-based CAPTCHA algorithm with amodal completion was proposed by our team. Our CAPTCHA causes computers a large amount of calculation costs while amodal completion helps human beings to recognize characters momentarily. Our CAPTCHA has evolved with aftereffects and combinations of complementary colors. We evaluated our CAPTCHA with deep learning which is attracting the most attention since deep learning is faster and more accurate than existing methods for recognition with computers. In this paper, we add jagged lines to edges of characters since edges are one of the most important parts for recognition in deep learning. In this paper, we also evaluate that how much the jagged lines decrease recognition of human beings and how much they prevent computers from the recognition. We confirm the effects of our method to deep learning.

Deep Learning Analysis of Amodal Completion CAPTCHA with Colors and Hidden Positions

Abstract: Existing CAPTCHA methods are able to be read by computers owing to the development of technologies We proposed a CAPTCHA method with amodal Completion in order to prevent computers from reading CAPTCHA We also applied aftereffects and colors for the method to enhance visibility However, our method was completely able to be read by computers with deep learning However, we were able to discover the undetermined element and problems in our research and the research was not a waste since we found problems and some hits to improve our method This study was conducted to expect that we get the opportunity to create a more robust CAPTCHA by improving them

Challenge of Deep Learning against CAPTCHA with Amodal Completion and Aftereffects by Colors

Abstract: We make experiments of machine learning for our CAPTCHA proposed as an effective CAPTCHA with amodal completion and aftereffects by colors. CAPTCHA is a method that distinguishes human beings from artificial intelligence in order to prohibit malicious programs from acquiring accounts on Internet. The most popular CAPTCHA is text-based CAPTCHA with distorted alphabets and numbers. However, it is known that all of text-based CAPTCHA algorithms can be analyzed by computers. In addition, too much distortion or noise prevents human beings from recognizing alphabets or numbers. As a solution of the problems, an effective text-based CAPTCHA algorithm with amodal completion was proposed by our team. Our CAPTCHA causes computers a large amount of calculation costs while amodal completion helps human beings to recognize characters momentarily. Our CAPTCHA has evolved with aftereffects and combinations of complementary colors. In this paper, we evaluate our CAPTCHA by machine learning since machine learning is faster and more accurate than existing calculations by a computer. We confirm the limitations of machine learning. Especially, we focus on whether a computer can recognize characters without knowledge of amodal completion.

Creation of Adversarial Examples with Keeping High Visual Performance

Abstract: The accuracy of the image classification by the convolutional neural network is exceeding the ability of human being and contributes to various fields. However, the improvement of the image recognition technology gives a great blow to security system with an image such as CAPTCHA. In particular, since the character string CAPTCHA has already added distortion and noise in order not to be read by the computer, it becomes a problem that the human readability is lowered. Adversarial examples is a technique to produce an image letting an image classification by the machine learning be wrong intentionally. The best feature of this technique is that when human beings compare the original image with the adversarial examples, they cannot understand the difference on appearance. However, Adversarial examples that is created with conventional FGSM cannot completely misclassify strong nonlinear networks like CNN. Osadchy et al. have researched to apply this adversarial examples to CAPTCHA and attempted to let CNN misclassify them. However, they could not let CNN misclassify character images. In this research, we propose a method to apply FGSM to the character string CAPTCHAs and to let CNN misclassified them.

A Machine Vision Attack Model on Image Based CAPTCHAs Challenge: Large Scale Evaluation

Abstract: Over the past decade, several public web services made an attempt to prevent automated scripts and exploitation by bots by interrogating a user to solve a Turing-test challenge (commonly known as a CAPTCHA) before using the service. A CAPTCHA is a cryptographic protocol whose underlying hardness assumption is based on an artificial intelligence problem. CAPTCHAs challenges rely on the problem of distinguishing images of living or non-living objects (a task that is easy for humans). User studies proves, it can be solved by humans 99.7% of the time in under 30 s while this task is difficult for machines. The security of image based CAPTCHAs challenge is based on the presumed difficulty of classifying CAPTCHAs database images automatically.

Strengthening Pix CAPTCHA Using Trainlm Function in Backpropagation

Abstract: CAPTCHA (Completely Automated Public Turing Test) security mechanism is a new and innovative technology in the world of the Internet and IOT. It is becoming a regular feature of the majority website’s login system. Pix CAPTCHA’s are more secure as compared to text-based CAPTCHA as it saves the websites from bots attacks. In this paper, we introduce another mechanism for the storage of Pix CAPTCHA using the techniques of artificial neural networks. The experiment results prove that the new model has a good recognition effect of Pix CAPTCHA when trained using backpropagation algorithm.

Challenge to Impede Deep Learning against CAPTCHA with Ergonomic Design

Abstract: Once we had tried to propose an unbreakable CAPTCHA and we reached a result that limitation of time is effect to prevent computers from recognizing characters accurately while computers can finally recognize all text-based CAPTCHA in unlimited time. One of the existing usual ways to prevent computers from recognizing characters is distortion, and adding noise is also effective for the prevention. However, these kinds of prevention also make recognition of characters by human beings difficult. As a solution of the problems, an effective text-based CAPTCHA algorithm with amodal completion was proposed by our team. Our CAPTCHA causes computers a large amount of calculation costs while amodal completion helps human beings to recognize characters momentarily. Our CAPTCHA has evolved with aftereffects and combinations of complementary colors. We evaluated our CAPTCHA with deep learning which is attracting the most attention since deep learning is faster and more accurate than existing methods for recognition with computers. In this paper, we add jagged lines to edges of characters since edges are one of the most important parts for recognition in deep learning. In this paper, we also evaluate that how much the jagged lines decrease recognition of human beings and how much they prevent computers from the recognition. We confirm the effects of our method to deep learning.