First, the basic knowledge of verification code
1. The main purpose of the verification code is to force human-computer interaction to withstand machine automation attacks.
2. Most of the verification code designers do not have the essentials, do not understand image processing, machine vision, pattern recognition, the basic concept of artificial intelligence.
3. The use of verification code, you can get rich, of course, to crime: for example, China Merchants Bank password only 6, verify the code form, the computer will soon be able to crack a rich account, many accounts can be online transactions.
4. There are also better design, such as Yahoo,google,microsoft and so on. And domestic Tencent Chinese verification code is difficult, but not good.
Second, artificial intelligence, pattern recognition, machine vision, the basic knowledge of image processing
1 Main flow:
For example, we want to identify the verification code from a picture, for example, we want to detect and identify a face from a photograph. What are the steps that might be taken?
1. Image acquisition: Verification code, directly through HTTP capture HTML, and then analyze the image of the URL, and then download save on it. If it is face detection and recognition, generally through the screen collection equipment, collected back, through a A/d turn operation, stored as a digital picture or frequent.
2. Preprocessing: Detection is the correct image format, converted to the appropriate format, compression, cut out ROI, remove noise, grayscale, convert color space these.
3. Detection: License plate detection and recognition system to find the approximate location of the license plate, face detection system to find all the faces of the picture (including the suspected face); Verification code recognition, mainly to find out the main areas of text.
4. Pretreatment: Face detection and recognition, the face in the recognition before making some corrections, such as the face outside the rotation, distortion and so on. I am here to identify the authentication code, "General" to do the text of the cut
5. Training: Through a variety of pattern recognition, machine learning algorithms, to select and train the appropriate number of training sets. Not as much as a training sample. After learning, the problem of poor generalization ability may appear here. This step is not necessary, and some recognition algorithms do not need to be trained.
6. Recognition: Enter the processing of the image to be identified, converted to the classifier required input format, and then through the output of the class and confidence, to determine what letter probably. Recognition is essentially classification.
2 Key Concepts:
Image processing: Generally refers to the digital image of some kind of mathematical processing. such as projection, passivation, sharpening, refinement, edge detection, binary, compression, various data transformations and so on.
1. Two value: The general picture is color, according to the degree of fidelity, may be many levels. In order to reduce computational complexity and facilitate subsequent processing, if the image can be processed into Black-and-white color without loss of critical information, it would be best.
2. Refinement: Find the skeleton of the image, the image line may be very wide, through the refinement will be the width of 1, some places may be greater than 1. Different refinement algorithms may have different differences, such as whether it is closer to the middle of the line, such as whether to maintain the Unicom line.
3. Edge detection: Mainly to understand the concept of edge. The edge is actually a dramatic change in image pixel attributes in the image. May be judged by a fixed threshold value, or it may be adaptive. Thresholds may be either global or local to the image. It must be good not to say that, but most of the time, the adaptive local threshold may be better. The analysis may be the color, or the grayscale of the grayscale image.
Machine Vision: The use of computer models to achieve human vision. such as object detection, positioning, identification. According to the level of the image understanding of the difference between the higher and lower levels of understanding.
Pattern recognition: A representation of things or phenomena (numeric, literal, what we want to say here is numeric), with some processing and analysis to describe, categorize, understand, interpret these things, phenomena and some sort of abstraction.
Artificial Intelligence: This concept is relatively wide, above all belong to artificial intelligence this big direction. Easy, don't overdo it. The academic understanding is to treat the problem of human's very "intelligent" things to the people who simulate to help the creature, especially in the computer.
Current 1/4 page
1234 Next read the full text
