Look at what the top US professors 'Computer Vision' course speak?2014 年 12 20 - 21 May, I was from Hong Kong to Shanghai, in accordance with plans to participate in the company's conference room, Los Angeles, UCLA professor of statistics from the University of California Alan Yuille (Alan Yule.) teaches a \u0026 ldquo; computer Visual forefront of curriculum \u0026 rdquo ;. Department of Computer Science from the major colleges and universities, Mathematics, Physics different backgrounds, more than 40 students to join. We exchange down very rewarding, but also very happy here and now. According to co-founder of Science and Technology Zhu Long (Leo Zhu) and mentor Professor Alan Yuille dialogue first question many people will ask, what is computer vision? Professor Alan first to tell you that this problem is actually eyes of the beholder, the wise see wisdom, and there is no fixed answer. With a smattering of Nike Air LeBron EE computer vision \u0026 ldquo; brick home \u0026 rdquo; angle, the author himself felt in computer vision is mainly used computers replace human eyes to see things. Professor Alan likes to use Aristotle's words \u0026 ldquo; To Know What is Where by Looking \u0026 rdquo; to interpret computer vision. To give you a better team of computer vision to have a concept, the author himself felt the best way was to name a few \u0026 ldquo; chestnuts \u0026 rdquo ;. Example 1 in the history of the World Cup soccer stadium, there have been a lot of \u0026 ldquo; the devil Goals \u0026 rdquo; mystery, in the end did not enter into the still, both players, referees, spectators are uncompromising, arguing for decades nor below. The most famous was undoubtedly the 1966 World Cup final, England's game against the Federal Republic of Germany, the UK team Hurst's shot landed just last gate line (http://songshuhui.net/archives/53739), The referee was scored, England With this goal win success. But for the ball in the end there was no crossing the line, the public has been debated. Until the 1990s, researchers at Oxford University's another way to answer this controversy, then share the video game computer vision techniques to calculate the position of the football in the air, and concluded that the year is likely to Hurst's goal invalid: the ball in the net from the recent time, there is 99.87% probability to become distance from the goal is greater than 6 cm. But this outstanding historical issues should not arise again in the future on the World Cup. On this year's World Cup in Brazil, \u0026 Air Jordan Outlet ldquo; goal-line technology \u0026 rdquo; was introduced to the match, soccer goal from close to the goal and then to pop up in the air throughout the entire detection and location can be calculated, whether the goal is no longer needs referee's eyes, the computer can tell you exactly. Rely mainly on two goal-line technology system, a system is a sensor system installed on the football inside and goal, the other one is set up in different areas of the field of high-speed cameras consisting of \u0026 ldquo; Women Lebron Hyperdunk 2013 \u0026 rdquo ;, Hawkeye system can provide a visual picture and time data analysis and calculation, marking the successful application of computer vision technology to the World Cup. Example 2 Another very common computer vision applications should be face recognition, we upload a group of photos on social networking sites after, you will find a few people above face was painted on the block, but this simple feature ten years ago, computer vision Nike Air Max 1 was a huge problem. The computer can not only face ring out, but can identify who this face is, and has been widely used in practice, between Hong Kong and Shenzhen several busy ports, face recognition function Lebron Slide 2 Elite has been self-clearance machines the use for many years. In the course of the organizers according to the company's products on display diagram, I try the company's recognition of the computer program for PK, see who can more clearly distinguish several Nike Air Max looks very similar to the photos found involved in attempts a dozen of the \u0026 ldquo; the ability to Nike Lebron 10.8 see the face \u0026 rdquo; no one can be more than the recognition success rate computers, as authors such as \u0026 ldquo; prosopagnosia \u0026 rdquo; people recognize the success rate is very worrying. Of course, in reality there are Nike Lebron X Elite many other applications of computer vision, such as the automatic recognition of Google's image search function and car license plate numbers, not list them here. In thirty or forty years ago, it used to be a computer vision we all feel very strange vocabulary, but now after Alan Yuille professors and other computer vision of the founders of the long-term unremitting efforts, computer vision technology has appeared in the World Cup, there has billions of dollars in the user's various social networking sites, it appears on the world's busiest port in the Customs \u0026 hellip; \u0026 hellip; we may cool for a variety of computer vision applications, wonderful products and a variety of science fiction future more interested, but these are inseparable from behind seem boring underlying theories and formulas. As a research scholar in computer vision, in the curriculum, Alan Yuille introduced to more cutting-edge computer vision theoretical research. Professor Alan Yuille first 30 years of research, the human visual fact in itself is a very complex thing in the field of computer Womens Nike Kobe VIII vision, machine and let people have the same level of visual perception is not easy. Alan tell you, open your eyes you see a thing, not even one second of time is irrelevant, seemingly did not bat an eye blink, his mind without thinking, effortlessly, but that does not mean that the visual The process is certainly a simple math problem, because the human brain is allocated a lot of resources for vision, all neurons in the 40-50% are associated with visual function related to vision than you think hard \u0026 ldquo; investment \u0026 rdquo; ratio of smell , hearing, taste is much more, this makes you see things easy. Moreover, computer vision, image itself is facing a very complex, an ordinary photograph a few megapixels, and each color pixel values have red-orange, yellow, green shades are many possible, so As a result, no two identical leaves in the world, there is no two exactly the same picture. The same person in different places, with different cameras, put a different pose, shot from different angles, shoot from different distances, wear different clothes shot, the resulting photos are different, but as a Get Info from the picture technology that requires different picture from those of the same person identified. With a photo of elephants, for example, the human eye can be whole elephant India into the eye, the eyes may be focused on a certain part of the body of an elephant, but it'll eyeball it. And for the computer, such as \u0026 ldquo; Mangrenmoxiang \u0026 rdquo; \u0026 ldquo; Benevolence \u0026 rdquo; as the extraction and processing of information image of each small part, or deal with the whole elephant information you need, how much computing time, the use of which algorithm and they are very different, the former is called Low-level vision, which is called the High-level vision what tasks to complete. Low-level vision can get rid of the elephant in the picture stains, you can identify the edges between the elephant and the lawn can be split from the photos about the two big elephants region, but can not identify what animal photographs; High-level vision contains a lot of artificial intelligence, such as it is just what the animals, elephant or Asian elephant African elephant, which sleep in the photo or in search of food, photos, there is no lack anything, and so each Species analytical reasoning and forecasts. Of course there is between Mid-level vision Low-level vision and High-level vision among some streaks such as elephants, the angle from which the photo was taken, how far from the lens elephants and so on. Can be found in an ordinary photograph contains \u0026 ldquo; \u0026 rdquo ;, a large amount of information on computer vision technology required to extract them all out, you need to complete the task is very heavy indeed very large, if the whole picture in all the information extracted simply require a Sherlock Holmes. Alan tell you that we can have a computer vision Turing test, that is, if one day, humans can see the computer system think of something, the camera and the computer component from a picture where you can also see all expect, both of there is no difference between, so that the computer vision in achieving the ultimate goal. Of course, this day is still very far away from us. The face of these difficulties, a few decades, researchers have proposed a variety of methods and theories, so that the computer vision field itself becomes increasingly large, and now I am afraid that no one book can fully encompass all knowledge of computer vision. After the introduction of the entire framework of computer vision after, Alan specific description of the various themes in the field of computer vision. Edge Detection (edge detection) and Image Segmentation (image segmentation) are the two most basic computer vision algorithms are all Low-level vision. Edge Detection very broad application, such as there is an edge between a face recognition time, face and background, eyes, mouth, ears, around the nose has an edge, if we can identify these lines from the picture , size and location information for each part of the face would be a lot easier. (Edge Detection) from the mathematical sense, means the edge of the color values of pixels in the picture there is a sudden change, of course, a lot of time edge is relatively vague, but the color value is not in place because of the noise and the edges will fluctuate, the need to find the edge pixels while avoiding the \u0026 ldquo; false edges \u0026 rdquo; identified. Now there are some simple and effective method for image Edge Detection Canny Edge Detector, Professor Alan methods on the basis of these statistics added elements: First, some pictures, the artificial line drawn all the edges, and then use these existing The correct answer test images on a variety of Edge Detector lot of testing, according to the statistics of the results, to better judge the edge pixels, equivalent to a variety of traditional methods once optimized. If a picture can be directly extracted from the edge, you can use these lines as a boundary map is divided into several parts, to face similar to the part extracted from the background image out of effect, which is the Image Segmentation. Of course, it can also independently of Image Segmentation Edge Detection completed, mainly rely on the consistency of the various parts of the pixels, such as a human face is the color of the skin, like the car surface is metal, the ground is the soil, the way to make a computer to \u0026 ldquo; Wumafenshi \u0026 rdquo ;. (Image Segmentaion: automatically put people in the picture, two different cars, as well as some background, each segmented) Following this line of thought, the easiest way is to SLIC (Simple Linear Iterative Clustering), equivalent to the picture The location adjacent pixels in accordance with similar \u0026 ldquo color values between each other; a feather flock together \u0026 rdquo; principle into several regions, the algorithm is very simple, mainly through continuous loop repeating simple steps. Alan also introduced a relatively more complex Graph Cut of Image Segmentation method, the user needs to do first is to draw a rough area in the picture (for example, to extract the face area, in the face of a rough draw around an area of the box), and then the program starts running Graph Cut the edge of the selected area will follow the \u0026 ldquo; energy minimization \u0026 rdquo; optimization, and finally converge on the edge of the area they want to divide. Alan display of Nike Kobe 9 Graph Cut four different optimization methods including Gibbs Sampling, Belief Propagation, Max flow / Min cut and MFT, and compared various methods Image Segmentation results obtained. At recess time, I also learned about the development of products according to company plans, and programs can write here some corresponding feelings and harvest, such as an automatic license plate recognition and brand (model) of the software, regardless of the car at high speed, or in the dark without lights at night, just under the surveillance camera coverage, license plate Nike Zoom Hyperflight PRM number and brand models can be easily identified, more interestingly, through photographs of the vehicle, can also determine the authenticity of the license plate (identify fake deck vehicles). In fact, computer vision, regardless of the license plate number, house number, or text field on the giant, identify typical tasks belong High-level vision text from a picture. Alan try to use more popular nowadays depth neural networks to solve this problem. \u0026 Ldquo; neural network Cheap Nike Free Outlet \u0026 rdquo; to mimic the human brain in the nervous system, the computer program in a virtual construct a model similar to the human brain of interconnected neurons. This model is like a machine, each neuron is a part, these parts are connected to each other, there will be very tight, and some will be very loose, the strength of each connection between a parameter. The depth of the neural network is a neural network in the most recent one kind of complicated. By giving ANN has identified a large number of images with text too, in the sample image and standard answers constantly \u0026 ldquo; Training \u0026 rdquo; below, the parameters changing the depth of neural networks, the final can naturally learn to recognize the wonderful picture The problem, as if the man himself does not know how his brain is working, we are not entirely clear is the depth of the neural network computer in the end how to complete the study. Alan said that the depth of the neural network consists of a number of layers of neurons, first from pictures to identify the relatively low-level vision of the information, and then drill down to extract the low-level vision of information. Depth neural network can fire up in recent years, mainly computer hardware computing capacity enhancement, such as GPU (video card) is widely used calculation, making this only \u0026 ldquo in the past; paper \u0026 rdquo; ideas can really achieve. The depth of the neural network to recognize text in the image recognition can be widely used in surveillance cameras and customs checks, and sometimes even exceed the accuracy of the ability to identify a real person behind dependent on a variety of recognition. During the course, Alan also introduced his research team presented Hierarchical Compositional Model for face recognition. In the model, data on the entire face is divided into three layers to deal with, from the highest level of the entire face to various parts Nike Air Griffey Max of the face (such as eyes), and then to all the details. This vision can be a model from High-level to Low-level vision different algorithms together. Throughout the course of the last, Alan shared his respect for what the students do research, how to write papers, reports and other aspects of how to do academic experience, and everyone conducted numerous interactive quiz. Some students choose what to ask Professor Alan PhD student standards, unlike popular belief low GPA, GRE scores and published papers, the professor Alan felt GPA only represent learning courses and test scores do not necessarily mean research high capacity to own his own example, course grades in school is not outstanding, but papers are often unable to determine the application of student papers in the end Nike Lunar Hyperdunk Low played Nike KD 7 Kids a big role. Professor Alan felt direct contact with the most reliable way, if there are undergraduate students participated in a summer internship in his lab, there are comparatively good performance, it is more likely to be admitted as his doctoral student, if there is no opportunity to direct contact, He will make reference to the recommendation of other researchers familiar, felt their views is more credible. Through this two-day course, the author received a rare opportunity, under the guidance of the world's top scientists in the field of computer vision, learn a lot of cutting-edge knowledge and latest research results on the field, thanked the TV again, thanks organizer. This article i dark horse All rights reserved Please indicate the source, tort reserved.