Soomal Insights 2018: Camera
Author:Soomal
Source:eng.soomal.com

Rate:00.00/0
Phone Camera: Speed Up

The advancement of mobile phone camera technology has never cooled down, but is still accelerating. Unlike the camera industry, which is more pleasing to the professional user preferences, the focus of the mobile phone camera field is how to make the mobile phone camera easier. In 2018, the camera and mobile phone finally went in two distinct directions, the camera industry. The heirloom has no value for mobile phone manufacturers.

Soomal believes that there are three technical keywords in the phone camera field in 2018: AI, HDR, and night scene.

AI

Sticking an AI tag is a trend in 2018, such as AI smart photo, AI portrait, AI beauty, AI photography wet, AI dream blur... Without AI seems not fashionable. Therefore, many of them are fake AI.

Many cameras have a scene recognition function, which calls different shooting parameters based on scene recognition, and then acts according to the routine. This process is not called AI on the camera, but on many mobile phones, it is called AI.

We think of AI, which requires depth-map-based image recognition and careful processing. Most mobile phones can only do is simple scene recognition, such as someone focus on flowers, and the phone will recognize flowers, then the corresponding strategy is to add saturation, but when the color itself is very bright, the colors will be overflow. For another example, face recognition, preferentially focus on the face and prioritize the metering, while if someone is recognized in the distant bush, and then the metering focus is not correct. This kind of low-level image recognition technology can only make rough corresponding processing. The so-called AI is more like artificial mental retardation, which not only improves efficiency, but also improves the failure rate.

But it is undeniable that AI is the general direction and is an indispensable part of future mobile phone imaging. Now, mental retardation does not mean that the future is not good. More and more portrait mode of cameras become clever, knowing that the skin retouching to distinguish between men and women and to distinguish skin color, but in the early years it will only whiten, and men will become a sisi.

Soomal believes that in the next few years, AI is the focus of mobile phone camera technology, it will have a profound impact on metering, focusing, image processing, etc., sooner or later will achieve qualitative changes.

HDR

HDR is the technical focus of the mobile camera in the last three or four years. After years of development, the HDR on the mobile phone has made great progress, and the camera HDR at this time is still in a relatively primitive state of technology.

HDR is based on multi-frame synthesis, and its difficulty is not in multiple frames, but in synthesis. Multi-frame mostly adopts the method of bracketing exposure. The frame exposed for a long time is mainly responsible for recording the dark part information, and the short exposure is responsible for recording the highlight information. The difference between the different frames is not the same, so the change of light and dark when synthesized into one frame It is difficult to smooth, so there will be gray cluster phenomenon, the tone becomes less natural, how to eliminate the gray cluster, this is the algorithm to solve. For example, if there is a moving object in the scene, or the picture is shaken, the picture cannot be completely aligned. If you want to do it, you must have an advanced merging algorithm. Some of the mismatched parts must be removed, but not everyone has a good algorithm. Therefore, some mobile phone backlighting scenes are prone to ghosting.

The benefits of HDR need not be said, but the negative effects of HDR are quite obvious. How to reduce the negative effect of HDR is the place where the HDR algorithm has been improving.

The iPhone XS/XR series, which were launched in 2018, use the newly developed Smart HDR technology to solve the problem of shooting fast moving objects. The overall tone control is more natural. vivo also released Super HDR in 2018, and for the first time applied to the X23, it achieved better detail in the dark part of HDR. Samsung's Note 9 also introduced "dark HDR", which can achieve HDR in more low-light scenes. In the past, HDR and multi-frame noise reduction could not always be achieved.

Since the development of the HDR algorithm, it can only be said that it is becoming more mature. There is still a long way to go from maturity. No HDR can cover all scenes. According to the current development speed, it takes at least two years to achieve the effect of basically no obvious negative effects. In the next two years, HDR will remain the technical focus that all mobile phone manufacturers must pay attention to.

Night Scenery

Brands=HUAWEI;Model=CLT-AL00;Focal Length=4mm;Focal Length In 35mm Format=27mm;F-number=F1.8;Metering Mode=Pattern;ISO Speed=ISO800;White Balance=Auto;Contrast=Normal;Saturation=Normal;Sharpness=Normal;Exposure Bias Value=0.0EV;Exposure Time=4000000000/1000000000seconds;Exposure Program=Normal program;Scene Capture Type=Standard;DateTime=2018.04.23 19:21:30
Brands=HUAWEI;Model=CLT-AL00;Focal Length=4mm;Focal Length In 35mm Format=27mm;F-number=F1.8;Metering Mode=Pattern;ISO Speed=ISO640;White Balance=Auto;Contrast=Normal;Saturation=Normal;Sharpness=Normal;Exposure Bias Value=0.0EV;Exposure Time=4000000000/1000000000seconds;Exposure Program=Normal program;Scene Capture Type=Standard;DateTime=2018.04.23 19:42:04

Huawei P20 Pro brings a "night scene" function, handheld 4 seconds for shooting low light scene, which can be said to be a sensation. By the end of 2018, many mobile phones have added similar functions, it can be said that this function has won the "tribute" of the peers.

Suppose that the scene we are in now is metering in normal mode. It takes 1/10 second @iso800 to shoot. At this time, we switch to the "night scene" mode, and take a 4 second shot and succeed.

Then, 4 seconds is 40 times the exposure time we actually need. Assuming that the 4 seconds have been continually framing the samples, then the final sample is finally synthesized, which is only 2.5% of the total sample. Assuming 4 seconds of continuous framing sampling, and using ISO 100 sensitivity, the final sample used is only 20% of the total sample. Regardless of how it is calculated, there are far more sampled samples than required. The popular multi-frame sampling method is that 100% of the samples are used for synthesis, and there is no redundancy.

Huawei's approach is innovative, which means that the algorithm can screen out enough high-quality samples for synthesis, such as removing ghosts caused by moving objects. This is to use longer time to improve frame quality, so the combined picture quality and success rate can be guaranteed.

We have written articles specifically about this feature. Welcome read.

The "Night Scenery" function is still not the best state. Therefore, Huawei is also constantly tapping the potential and updating the algorithm. It is only fortunate that the new algorithm has more problems. If you look farther, the "Night Scenery" function will definitely get more improvement, and sooner or later it will be integrated into the normal shooting mode.

Calculating Photography

AI, HDR, night scene, simulate blurring, focus logic optimization, etc., make every photo taken by the phones process different from that of cameras. The algorithm not only controls the image quality, but also controls the behavior of the camera. Every second of the viewfinder is on, the phone will constantly calculating, judging the user's intention, finding the focus object, matching the scene parameters, etc. After touching the shutter, it will align the multi-frame image, noise reduction, and even Use the neural network to imitate the human brain to judge the image and so on and so on. Apple officially has a slogan " A picture is worth a trillion operations.", which may be exaggerated, but the operation is indeed all the time.

This means that mainstream mobile phones have entered the era of Computational Photography, and cameras are still very primitive in these technologies. In the era of computational photography, algorithms become more and more important, and algorithms are the basis for good or bad cameras.

The difference between computational photography and traditional imaging methods is that computational photography is close to the average consumer, and some control techniques and primary experience are done by the system, which is better than ordinary rookie. Soomal believes that computing photography does not reduce the fun of photography. The core experience of photography is to capture and compose images rather than fiddling with machines.

Please Rate 1 2 3 4 5 6 7 8 9 10
Soomal.com