Home > News content

Microsoft self: let you save Black & not take pictures

via:博客园     time:2015/12/29 21:00:54     readed:1965

You know, not long ago released Taobao consumer data report, self timer lever of the main force in the purchase is actually 50-59 years old crowd! No matter whether you bought self rod, no doubt, now self absolutely not only belongs to the youth entertainment.

In the United States and the United States led photo the sun before social media, most people have an indispensable step, that is to use the retouching software on the picture of landscaping. Currently on the market a wide variety of pictures to beautify the software, operation is complicated and changeable, accidentally put themselves into P

Microsoft's latest picture landscaping applications

Self beautification, just a key

So if you want to write a Raiders use it to Microsoft customers, it ought to be: 1, open the Microsoft self application; 2, in front of the camera pose expression; 3, the pictures, then there is no then

However, if you write a technical process for Microsoft self, it is very complicated

Self timer process

When you open the Microsoft customers ready to begin taking pictures, the application will be active sensing the surrounding environment, if the light is too dark so that a camera can produce noise, such as night outdoor parties, bars and other environmental, it will automatically open noise reduction mode, to ensure that the generated picture clean and free of noise.

Whether through Microsoft led to shoot photos or import images from the album, it will in next to 1 second of time, on the input images for fast processing, mainly includingNatural beauty,Intelligent noise reductionandExposure enhancementThree parts, for both auto heterodyne photograph, or inhuman picture (such as landscapes, objects, etc.) can achieve the ideal photo beautification effect.

Natural beauty:

In fact, different gender and different age of the user for the self retouching requirements vary greatly. For example, a family of three generations of the photo, grandmother, mother and daughter three generations of the face have been treated in the same way

Pictures from the Internet

Microsoft, LED intelligent beauty function in beautifying portrait photo, automatic identification picture portrait of the age, gender, skin color, and light and other factors, and based on the corresponding to eliminate users do not need the wrinkles, bags under the eyes, freckles, and acne.. But for users who wish to retain the physical characteristics, such as hair, tattoos, bushy beard and or facial ornaments, such as earrings, nose rings, glasses, etc., the application will keep changeless as far as possible, eventually make the whole face beautifying effect is more natural.

So both single auto or photo, a key natural beauty can let photos of all people have the most suitable for their own beauty effect!

Such a simple and general beauty behind the operation, technology is not easy. First of all, the system should face detection JDA (Face (detection) on the photoJoint Cascade Face Detection and AlignmentECCV 2014), to detect whether a human face exists in the picture, a total of how many faces, and the face appears somewhere in the picture.

Then, using the feature point detection technology analysis of the different features of each face to what extent the beautification system. Microsoft self timer is used in Microsoft Asia Research Institute in 2014 CVPR conference papers:Face Alignment at FPS via Regressing Local Binary Features 300068 face feature points in the proposed detection method. By this method, the system can detect the facial feature points quickly and accurately and. Next, based on the smart way to high-rise, the algorithm to extract information from the face in the picture, including age and gender discrimination (Blessing of Dimisionality: High Dimensional Feature and Its Efficient Compression for Face VerificationCVPR 2013). This technique is also used in a well-known How-old.net (Microsoft Yan on the age of the robot).

Finally, the application of the

Intelligent noise reduction:

Intelligent noise reduction function is a self Microsoft

And Microsoft has released the time-lapse photography applications hyperlapse similar Microsoft customers include fast inter frame alignment and multi frame noise reduction technology in the key technology of the intelligent noise reduction function. A single photograph of noise is around the average value of fluctuations up and down, and if the two picture of the noise superimposed on the words, you can derive a more clean images. The increase in the number of pictures in the photo, the noise will gradually decline. Based on this principle, researchers have suggested that the a fast multi frame noise reduction method, related papers have been published in the last year of the siggraph Asia Conference (Burst Images Denoising FastAsia SIGGRAPH 2014).

But for the photos of tremor phenomenon, if only simply to average superposition of multiple photographs of the burst, it will produce ghosting effect. So that a fellow between the frames and the frames of the camera model estimation method, not through the sensor but completely through digital frame alignment method for processing. Secondly, when the frame and frame alignment, because of different depth in the scene, a tiny movement will lead to the same pixels in different frames corresponding to the object is not the same, so if just do a simple picture mean words, there are still may lead to picture ghosting. So algorithm by the researchers will screening selection consistent pixel points brought average, the average is not only occurred in the time domain also occurs in the spatial domain, so that the noise reduction algorithm for both photo face or non face images have a very ideal effect.

Exposure enhancement:

I believe that all met wants to shoot the scene is limited to angle, light, and other problems resulting in backlight shooting, and Microsoft led auto exposure function can photos for regional exposure adjustment. For example a backlit photos, and face all backlit scenes, relative than the background is very dim. Or is the face of the sun's photo, the light is very bright, and it is difficult to obtain the natural expression of face. In this case, to shoot a picture for post processing, the traditional repair application diagram usually global adjustment frame in the light, that will be the prospects appear pulls bright at the same time, the background light is too bright, even exposure.

Faced with this situation, Microsoft will automatically self regional frame exposure analysis and adjustment. First system will automatically detect the backlight, and for each region provides optimal local exposure compensation, avoid the whole picture in picture over exposure, and ensure that the photos to see backlighting of the scene at the same time to maintain the photographs of the administrative levels sense (Automatic Exposure Correction of Consumer PhotographsECCV 2012).

The scenery mist common, researchers have also developed to fog algorithm (Single Image Haze Removal using Dark Channel PriorCVPR 2009). The algorithm with exposure correction algorithm fusion to enhance the image resolution, so as to achieve the use of ultraviolet (UV) filter or circular polarizer (CPL) professional photo effects. Of course, when PM2.5 burst table or stay indoors or take a mask, don't take pictures

Light applications, big technology

From the prototype to the application, Microsoft Asia Research Institute Visual Computing Group for the whole Microsoft self provided technical support. Visual computing research direction and professional group covers the entire category in the field of computer vision, including face, image editing and computational photography, image recognition and detection.

The face research includes face (feature points) detection, face recognition, face recognition and so on property. The research results has in Oxford, Microsoft plans face recognition API API (face), windows Hello brush face boot, boot Kinect identification authentication and Microsoft wheatgrass and other many and face related products and services.

In image editing and calculating the direction of the photography, visual computing group research mainly automatic image editing and camera enhancement technology and so on. In December this year, Microsoft Oxford plans to just open the video API (Video API) is also the research direction of the figure. Up to in addition, OfficeLens, hyperlapse users such as tens of millions of applications can also see visual computing group in image editing of research results.

In the image recognition and detection, the research group in terms of image content understanding, recognition, detection and segmentation have made breakthrough progress. Not long ago, held in the world's top universities and research institutions2015 ImageNetComputer vision recognition challenge, the group of researchers with the latest breakthrough in deep neural network technology, with an absolute advantage won the image classification, image positioning and image detection are three main project champion, defeated from academic circles, enterprises and research institutions, numerous contestants, to create the world's most accurate computer vision system.

When talking about Microsoft led this one app behind the technology demands, Microsoft Asia Research Institute of visual computing group in charge of researcher to the yuan Rd said:

Let technology tend to be invisible

In the Microsoft released many familiar with the products and services, can see from Microsoft Asia Research Institute of the cumulative results of basic research. Microsoft Asia Research Institute is like the engine of innovation, continuously the cutting-edge scientific research achievements transformation to Microsoft products and services. Ordinary users do not see behind the complicated technology operation, can truly enjoy the convenience of technology.

Referring to Microsoft after the self timer technology, Yuan Road mentioned three development direction. The first point is more intelligent image processing and camera enhancement, such as increasing the intelligent classification of the scene to allow mobile phone automatically select more suitable for processing and shooting mode. The second is to increase face more editing effects, so that users can more personalized customize different modelling, increase more alternative facial enhancement effect, such as the choice of a face lift with the not. The last point is the technology used to enhance the image quality of intelligent real-time processing in video. The final goal is to these complex technology applied to people's life, let everyone can really feel the intelligent and personalized computing experience.

At present, Microsoft has self application log in the iOS store (download link:Https://appsto.re/cn/UTAD_.i), for everyone to download. In addition, the micro channel of Microsoft Asia Pacific R & D group innovation incubator group public No.Http://selfie.microsoftcafe.net/Can achieve some of the features of the application, go try it!

For more information, please scan the QR code below. .

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments