Baidu\'s \"funny face\" behind what is black technology?

Editor's Note: the author of Gong Yazhuo, poly-founder of Rainbow optical, Shanghai Jiao Tong University, Dr. Biometrics research, especially in the field of Iris recognition has over 15 years of research.On September 8, 2015, by many geeks dubbed the "world's raffish" "Baidu deep learning laboratory", was officially released worldwide Face funny face APP–face It.On this APP, user self portrait or upload any photos of faces, you can automatically generate materials. "The face" can be either using the front-facing camera on the material attached to the user's face, real-time interpretation; you can also use a rear-facing camera, different faces and posters, film and TV picture fitting, no traces of PS. By "click Photo" or "recording 20 second video" share to micro-credit, micro-blogging, Qzone. VERUS iPhone 6 Case"No surgery no scars, face a key big Lady" behind the face from this interesting application, is face recognition technology.Face recognition technology, consists of three key components: detection, feature extraction, matching face recognition.1, the background scenery for face detectionIn other words, face first to determine user uploaded photos in there face, if there are, then use left-right symmetry of the face position face and determined by characteristics such as skin color, eye color, hair race.2, to preprocess the face, extraction of facial featureOnce a face is defined, it can spread from the center using human characteristics that the gray complexity, depth and matching algorithm to detect a number of key points. VERUS iPhone 6 Case (Face: cute dog)Face "image mosaic" function is to find key points such as the mouth, nose, eyes, eyebrows, and integrate the user selected picture material. Key points of human face can be divided into different classes, such as 5, 252, which correspond to different adjustment needs. Under normal circumstances, eyes, nose, mouth on both sides which is detected 5 key points, you can do step 1 locate the face. (Tom Cruise 5 facial key points)If 25 people face a critical point in entertainment applications can be adjusted for more facial details, such as artifact "Mito" familiar "faces with big eyes" aesthetic, lock eyes, mouth, jaw and other key points, female goddess in one second.2015 hot global Microsoft "guess the age How-Old.net", are based on the face with age 27 distinct facial features, such as eyes, mouth, nose, and other key points, determine the user's gender and age. "Time is butcher knife" as time goes by, abrasion, the skin appears dark brown spots and wrinkles, drooping eyes and mouth will, age is the key factors affecting the accuracy of face recognition. (25 key points corresponding to the face area)3, matching recognitionFace recognition (Face Identification) is the identification of human faces in a database of known face match, determine the user's identity. Domestic matchmaking jiayuan, using facial recognition technology, from massive amounts of data screened most similar face with users to upload photos for single men and women "matchmaker", in essence, is what a young man with "husband". Baidu's magic map, has launched a new HTML5 games, user self portrait or upload photos, games will tell you which stars and is similar to how much similarity percentage.Face recognition technology for "marriage", "celebrity" this type of entertainment applications for recognition accuracy is not high, it can meet the requirements of. However, once used for high security level and accuracy of face recognition is not ideal. The most dramatic case is that Australia Sydney Airport was set to have the highest accuracy of face recognition systems, hopes to identify terrorists in passengers, and the precision of the face recognition system is disappointing, drove the United States famous actress mistook Winona-Winona Ryder Reid Genben-Osama bin Laden Osama bin Laden of terrorism.(Face recognition errors: Winona-Rhett did bin Laden really like? ）Spending huge sums to purchase face-recognition system, why I made such a stupid mistake?The most important reason is that face "of non-rigid" feature."Non-rigid" mainly for the face image lighting, posture, facial expressions and other differences under the different conditions, hair, beards, makeup, jewelry, clothing changes, and the impact of factors such as age, disease, injury. Congenital instability of facial features, along with the changing external conditions, this is leading to root causes of face recognition accuracy is not high.In high security applications, the face recognition technology is not used independently, often used in conjunction with the other biometric recognition more accurate, for multimodal recognition. The most common multi-modal combinations, that is, "face + IRIS". "Face + IRIS" multimodal recognition system"People face + IRIS" of more die State recognition system, General using process is: used monitoring camera on from people for precision lower of people face recognition (errors general rate about 1/1000), once found and police provides of suspects facial similar of people, will will issued alert, again used precision more high of Iris recognition (errors general rate about 1/1 million), further precision judge criminals of identity.Faces can be collected without user matching case photos, little invasive user experience, making it "has the highest ease of use" biometric and IRIS, is recognized as having "high precision" biometrics. Therefore, the "face + IRIS" multimodal applications, the advantages of both (j. w. Cowie, high accuracy) set in one, is the current biometric applications in the most advantageous technical solution.3701 votesiPhone 6s Plus\rIn actual use, Plus feel just like an iPad Mini–, especially in landscape mode, Plus cross-screen keyboard or directly join the cut, copy, paste, and the arrow keys, and then create a new note and email are suspended like a piece of paper on top of the entire interface, very unique. It seems that Apple is great at screens of different sizes have different characteristics of the interface.\rView details of the voting »