Mosaic-based graphical man-machine system design and analysis of recognition

Paper Keywords: denial of service attacks demonstrate interactive robot mosaic of Information security
Abstract: The robot can continue to access the site resources to consumption of resources, resulting in denial of service attacks. To solve this problem, Basso, who presented a mosaic of human-Computer interaction based on proven algorithms. Although the algorithm to a certain extent, prevent denial of service attacks, but there are also disadvantages: the synthesis of dance films there is always a true picture is not completely covered, and overlap each overlap are only 1 / 4 part. The arrangement really makes the picture very regular, really easy to picture the location of leaks. For these shortcomings, an improved human-Computer interaction based on the mosaic show that the algorithm, to prevent denial of service attacks.

Information security, including confidentiality, integrity, availability, non-repudiation, and several other aspects of the basic properties. Denial of Service (denialofSelW'1oe, referred to as DOS) attack is a destructive attacks on availability, it is mainly by sending packets over request, the consumption of network bandwidth or system resources, so that overloading the network or system services, resulting in decreased quality of service, and even paralysis or stop. DoS attacks easy to implement, difficult to prevent, Information security has become one of the hot areas of Research.

l knowledge.

1.1 The proof of human-Computer interaction
CAPTCHA fCompletelyAutomatedPublicTuringTesttoTellComputersandHumansApart1 means the automatic distinction between Computer and human Turing test, also known as automated Turing test l] 1. It is different from the traditional Turing test, but by the computer to distinguish the user is human or machine.

A typical CAP1 �� HA must have the following attributes:.

1) For the human user, should be fast and easy:
2) should accept all of the human user, without any discrimination,
3) almost no robot can solve this problem,
4) Even if that algorithm and data, can also resist the attack.

CAPTCHA can be divided into: character recognition, image recognition and speech recognition.

CA CHA text-based test using a computer program embedded in the image which does not recognize the extremely distorted and corrupted text. These pictures are usually easy to read for humans, but it is usually the automatic procedure is difficult to identify, even with the best optical character recognition software, text-based CAPTCHA must resist "part than the recognition of" attacks I.

Image-based CAPTCHA test requires users to solve the problem of visual pattern recognition or understanding of the concepts expressed by pictures. Since the difficulties of displaying images require a higher and larger area, resulting in the server burden. In addition, a large database, there may be problems such as with a particular site does not match the theme.

CAPTCHA-based audio tests focused on the machine difficult to understand distorted, degradation and background noise of the oral [61, vulnerable to outside interference and reduce the degree of speech recognition.

1.2 Mosaic based recognition algorithms graphic design and analysis of human
With the computer vision and pattern recognition advances, this text-based CAPTCHA is not the original so effective, more vulnerable to specific attacks. At present, the computer can not do a lot with the vision-related tasks, and these tasks is for humans easy, you can use this to design a method to distinguish between humans and robots.

AlessandroBasso and StefanoSieeo proposed an algorithm named "MosaHIP" is the Mosaic-basedttumanInteractiveProof (proof of man-machine interaction based on Mosaic) 17], which uses the existing computer in the implementation of some of the problems: 1) in a complex background Next, in the region of interest in image segmentation:
2) The chaotic situation in the background, the identification of specific concepts:
3) The specific transformation is applied to the image after the graphic match.

"MosaHIP" is based on the image for patchwork, mosaic, using a number of small, partially overlapping images to form the idea of ​​large images. These small images from two different categories:
1) describe the real, meaningful concept images:
2) Describe the concept of artificial or meaningless images.

Only a small part of the image is true, the first category, they need to be user identification, they are placed in pseudo-random patchwork of images. And overlap each other, which, for the computer, it is not easy to identify them. The rest of the image is really false color image color histogram of the random generated graphics, lines, it is used confusion in the background, purpose is to enable the robot image is really difficult to identify.

Figure 1 shows the concept-based MosaHIP, requires the user to identify images to be put together to include the true image. Net 2 shows the "top" of the MosaHIP. User identification requirements described in "some presence" meaning and "placed in the top", and is not covered by other images images.


AlessandroBasso and StefanoSicco the proposed algorithm can to some extent, although the distinction between robots and humans, but there are some drawbacks:
First, the description of the algorithm there is an error, in the fourth step, to determine the region where the image I_l will I_] is divided into four regions the same size, randomly select a region, and placed Ij in this area, and Ij an overlap, which there is 1 / 4 chance to be completely covered in the free paper image Ii Links Download Center http://www.hi138.com addition, the algorithm in terms of security may also exist some shortcomings: 1) whether is based on the concept or "top" of the MosaHIP, have not been covered by a true picture exposure, 2) each place is really a new image, it will cover the area on a picture of the 1,4 , no matter what the order really pictures, location, what they are really pictures of the area where the image location is there is a pattern, which makes it easy to picture the robot recognize
2, an improved human-computer graphics-based recognition algorithm mosaic
AlessandroBasso and Stefa130Sicco of improvement on the proposed algorithm: 1) P in the image database, select n images, add them to the real image set R = {I .... , 1, if the method is based on the concept of random selection of images I �� R, to determine its species do not G, or else I: = I.2) For each Ij �� R, randomly selected scaling factor Si, S �� Sj �� S, according to the Ij Sj scaling function to use to determine I. The need for rotating, random selection of rotation 0,0 �� o �� 0, according to 0. Spin Ij, Ij determine the transparency of the transparency factor of random selection, 0 �� T �� TAX, if the T. :! = O, Ij use based on the transparency of function:
3) Make a long transparent n m wide image c: e at random in the choice of a position, I �� R placed in this position, make sure I did not exceed the boundaries of c, if Ii = I, D, preserved in the collection I, the upper left and lower right coordinates:
4) For each IiER, 2 �� j �� n, determine a region where the image Ij, Ij a will. Is divided into four regions the same size, randomly select a region, according to the direction of the region, in the region random selection of a new region, this new area of ​​the region> = original area of ​​3 / 5, and placed Ij in this new area, and the li of a duplication and ensure that it does not cover the Ij. ,, If the method is based on the concept of pre-placed Ii can not overlap I, 1 �� k �� j, if Ii = I, I kept in the collection D, the upper left and lower right coordinates
5) c of the color image histogram calculation, hist (c), making long-m wide n the background image b, in the collection RGBh and RGBI color formation in the random selection of a color gradient to fill b, RGBh including his @) in the k most common colors. RGB1 including the rest of the color,
6), and really create a false image area similar to the image f, in the collection RGBh and RGB1 choose a color in the random formation of a color gradient to fill the f, f used in the color RGBh a variety of graphics and paintings line, the color changes with RGB1 f the pixel color in some regions, if it is based on the concept of method, the same area I divided into 4 areas, random selection of a region, and place f in this area, and I . overlap and to ensure that images do not completely cover the image f I, if it is "top" approach, will I be divided into four regions the same size, random selection of a region, and place f in this area, and I overlap and to ensure full coverage of non-picture image f I,
7) continued to produce images f, will be added to the background image f b, from top left, repeat step 7 until b is false image completely covered, *
8) The use of Floyd-Steinberg Dithering Algorithm reduced the number of colors the image b,
9) the synthesis of images overlapping the background image c to b, because c has a transparent background, and now includes true and false b images
lO) using the deformation function of b (with not a random selection of input parameters),
l1) Returns a collection of images b and the coordinates of D, if the method is based on the concept, but also to return kind G.

The algorithm in step 4) and 6) improve, at 4), this improved algorithm makes a true picture of each are covered with a really random picture of the area, but does not interfere with human identification, in the 6), did not cover the true picture, the improved algorithm to use a fake picture and this picture really overlap, thereby enhancing the security of the algorithm
3 Conclusion
This paper analyzes the harm and denial of service attack defense method, describes the status of human identification and recognition associated with the human knowledge of a Turing machine and the Turing Test and so on, on this basis, this paper based on Basso and others Mosaic graphical man-machine recognition algorithms, and analysis of the algorithm and found that the algorithm description and safety are flawed and inadequate. For these deficiencies, the paper Basso et al mosaic-based human-computer graphics based on the recognition algorithm has been improved to enhance the security of the algorithm. Links http://www.hi138.com Research Papers Download