Night scapes beautifully captured on paper: "Train-Buff" professor at ICT develops technology

Even if you take a beautiful nighttime photograph, you cannot print it well on paper. Professor Sakuichi Ohtsuka, currently a professor at Kanazawa Institute of Technology, and his research group at the International College of Technology (ICT) in Kanazawa City developed technology to solve this problem and print nightscapes exactly as they look to the human eye. Ohtsuka has been a "train buff" (train photographer) since he was a child, and he loves taking photos of trains. However, beautiful and emotional night views look dull when they become photographs. From the simple question, "Why is this happening?", he has continued his research for nearly 50 years to solve this problem. "My hobby and practical benefit came together," says Ohtsuka with a smile. Not like a railroad track, his research goes on and on forever.

What a steam locomotive taught him in childhood

Ohtsuka was born in Tsuyama City, Okayama Prefecture, where his parents' home was located. After that, he often traveled back and forth between Ishikawa Prefecture and Okayama Prefecture by train to see his parents' grandmother, and a steam locomotive (SL) sometimes stopped right in front of him. He was fascinated by the power of the large, shiny black body and the grand sound of the whistle.

"I have been a fan of trains since I was a child, and in junior high and high school, I recorded the sounds of SL trains on cassette tapes," Ohtsuka reflected. During his high school and university days in the late 1970s, he chased the last active steam locomotives and traveled as far as Kyushu and Hokkaido.

Photography methods changed with the times, moving from film cameras to digital cameras. In the early 2000s, camera sensors became more sensitive, making it possible to take photos of night views. Around this time, trains also became colorful, and Ohtsuka took and printed photos, thinking "I want to record the beautiful appearances of various trains."

Night trains are beautiful because light reflects on their jet-black bodies in the dark night. "Being dark" and "being black" are completely different, and he looked for photos that could tell these two apart. However, the impression of the printed photo was very different from the real thing seen with the eyes. "Why is there not much difference during the day, but so much difference at night?" This question was the "starting station" of his research.

Doubting textbooks and making hypotheses on the functions of the brain and retina

After working for a private company, he entered the world of research. His specialty is visual information processing. While studying human interfaces for displays and how things look in bright and dark places, a question came to his mind regarding a statement written in textbooks about the eye. The text said, "Light entering the eyeball is converted into nerve pulses by the retina, and then information processing is performed in the primary visual cortex and later parts of the brain."

In this context, the retina is treated as a simple signal converter, just like a camera's sensor. It is known anatomically that after the visual cortex, information is broken down into various components and then put back together. After local processing is done in each organ of the brain with different roles, the components are integrated and recognized as vision, though the detailed mechanism is not well understood.

According to Ohtsuka, even if you move your eyes quickly from one place to after another, the brightness tone of the whole field of view stays stable. If the conventional theory that "the brain processes vision only inside the brain" is true, "the brain would have to do many high-speed processes based on local processes to show the image." "To use an analogy, it would be like multiple NVIDIA GPU semiconductors running at full power without generating heat. Could that really happen?" He felt that something was wrong.

He also thought about the textbook statement that "the suprachiasmatic nucleus (SCN) in the hypothalamus of the brain senses light information and adjusts the body clock." He felt, "It is somewhat unnatural that the suprachiasmatic nucleus focuses entirely on non-visual body management and has nothing to do with how things look."

Therefore, Ohtsuka made a hypothesis: "The retina works together with the suprachiasmatic nucleus to act as a highly functional active sensor. The conventional theory that 'the suprachiasmatic nucleus affects only non-visual physiological phenomena such as sensing blue light or sleeplessness' is partly correct and partly wrong. It does not just feel non-visual information, but it has a role to adjust how we feel brightness according to the day and night brightness, and it works even in the early stages of vision."

The human eye: Strangely more powerful in dark nights

To verify that this hypothesis was correct, a steady experiment like a "local train that stops at every station" began. First, he studied deeply what it means to "see." Research on vision, like research on hearing or touch, makes it difficult to match your own senses with another person's senses. Therefore, he decided to show "how dark it can be for people to see things, or how much brightness is needed to see things" using the human eye and various images from cameras.

The moon has been an important object of observation for humans since ancient times. As seen in Waka poems and songs about looking at the moon, people have looked at the moon with the naked eye since ancient times. At the same time, the moon is a light source at night, and if there is moonlight, people can walk into quite dark places.

On the other hand, during the day, you can put the sun (the light source) at the edge of your view, but you cannot look directly at it. Because you do not look at the bright sun, you do not need to recognize the difference between light and dark as much as at night. It seems like we recognize the difference between light and dark the most during the day, but actually, we do not.

The range of brightness in a landscape from the darkest part to the brightest part, or the range of brightness that the human eye can recognize or a camera can shoot, is called the "dynamic range." For eyes and cameras, a larger number means better performance. The human eye has a dynamic range of 30,000 to 1 during the day, but at night, it is highly powerful at over 1,000,000 to 1.

Because of this, when your eyes get used to the dark, you can identify objects even without streetlights. Bright parts, like the lights of a distant city, traffic signals, and the colors of stars in the sky, look colorful just like during the day, while dark parts become close to monotone. This is called "mesopic vision." Two types of photoreceptor cells, which are cone cells that feel color and rod cells that do not feel color, work at the same time. This lets us identify colors from evening to night, even when bright places and dark places are mixed together.

The difference between the "camera eye" and the human eye

The camera images we usually see are called SDR (Standard Dynamic Range) images.In the photographs taken at the Nagaoka Fireworks Festival in Niigata, Fig. C is the SDR image.The dynamic range is about 200 to 1, which can be displayed on a screen without problems and can usually be printed on paper too.

On the other hand, the dynamic range of our daily environment is much higher than SDR, and an image that records this directly is called an HDR (High Dynamic Range) image. In the fireworks photos, Fig. A corresponds to this. Even if HDR images are saved correctly as data, they cannot be printed as they are.

To overcome this problem, two main methods are used. The first method uses extra equipment like reflectors or lights during shooting to change the landscape (the subject) itself so that it fits into a normal SDR range. If you can use extra equipment properly, you can get a beautiful image.

However, if it is difficult to change the landscape, the camera can only record a small part of the brightness information properly. In the example of the fireworks photo, the normal SDR image (Fig. C) has its bright parts blown out into white, and its dark parts are crushed and become invisible.

The second method is an image compression technology called "HDR toning," which is standard in recent digital cameras and smartphones (Fig. B in the fireworks photos). Using the function of the human visual cortex as a hint, this method changes the image boldly, focusing on local processing. It emphasizes the color, vividness, and contrast of both the bright and dark parts to fix the weaknesses of conventional SDR.

However, even with this technology, a scene like fireworks results in an image that is clearly different from how humans see it, as shown in Fig. B. In this way, we can say that the conventional "camera eye" looked at things with a mechanism different from the human eye. Even though both are called "eyes," they were different, like a "double-track railroad line."

Can image processing reproduce "optical illusions"?

It is difficult to print HDR images on current plain paper. To print a "beautiful photo just as it was seen," we must improve the technology that brings the camera's way of seeing closer to the human eye's way of seeing.

For example, to see the landscape in front of us correctly, we need to have an optical illusion. An optical illusion is a phenomenon where the brain recognizes a shape or size different from reality when processing visual information from the eyes. Human eyes are excellent. They compress brightness, lift dark parts, and let us feel light naturally. In other words, humans do not see the actual landscape, but see an image adjusted by the eyes and brain and think that it is a natural landscape.

Therefore, reproducing data recorded in a camera on paper does not look beautiful. In short, if we can reproduce human optical illusions through image processing, we can print beautiful images on paper. Finally, a light appeared at the end of the "tunnel."

Ohtsuka studied how much adjustment is needed to get closer to the human eye using various dynamic ranges. The number of images was 30,000. "It was a huge amount of data, like what an AI handles. I kept studying how the eye sees inside a black box that cannot be seen," he looks back on the laborious work.

Fireworks photo: Two-step processing makes it "look real"

Around that time, luck was on his side. At the Nagaoka Fireworks Festival in Niigata, where the home of his wife's parents is located, he won the lottery for the photographer seats on the riverbed of the Shinano River. In August 2025, he kept pressing the camera button at the perfect photo opportunities. Using those images, he tried processing them with numbers calculated from the huge database.

First, he compressed the brightness of the HDR photo to match the human eye. He processed it to lower parts that were too bright and lift parts that were too dark. He successfully compressed the excessively wide brightness range of HDR down to 70 to 1.

At the same time as this processing, to make it fit paper, he did extra processing to keep important parts and close unnecessary gaps. When he adjusted the image to look natural even at 30 to 1, it became possible to print it on paper. Through this two-step processing, "matching the human eye" and "matching the paper," he succeeded in reproducing beautiful night views and detailed day views on paper. This is the hMDR (hyper-realistic Moderate Dynamic Range) image.

He took photos of night views at Kenrokuen Garden in Kanazawa, which is close to Professor Ohtsuka's school, and the plaza in front of JR Sapporo Station, and printed them so they looked "exactly as they were." He estimated that the reason these night images look natural even during the day is that "the suprachiasmatic nucleus controls the function of the retina and automatically converts the vision to match the situation." As a result of testing many different scenes, he no longer found any cases that contradicted his hypothesis from an image processing point of view. The research he continued for a long time finally arrived at the "terminal station."

A future helping criminal investigations and more

However, Ohtsuka does not intend to "go into the train depot" as an "out-of-service train." He plans to "transfer trains" and continue his next research. "I want to apply this to images that need clearness, like disaster prevention cameras, street security cameras, and drive recorders, and make it popular," he spoke of his practical goals.

In addition, he said, "Right now, the verification from the viewpoint of image processing is almost finished, but we need biological proof that the actual functions of the retina and suprachiasmatic nucleus match the hypothesis. This is impossible with our research setup, so we need to do interdisciplinary research and check it from multiple viewpoints through animal testing. In reality, I think it will take a long time, on the scale of decades." Ohtsuka has also improved the technology introduced this time. He developed a technology to print nice photos that look natural even under bad HDR environments, such as indoor backlight during the day, which was difficult before. He presented his latest results at an international conference in Los Angeles, USA, in May.

Ohtsuka says his research "grew out of his hobby," but it has great social meaning. Among the cases investigated by the police, the percentage solved by images from security cameras is 17.6% (2024 Police White Paper) and is increasing year by year. For a safe society, we should say, "Let's get going!"