These days, most of us use modern digital technologies, ranging from cell phones, digital camera, GPS to Internet apps like Facebook or Twitter. Digital technologies has revolutionized how we interact with information, how we communicate and how we manipulate and store the data. Anyone who has a digital camera (and many of you probably own more than one) must have tinkered with image processing tools. Some of you did enhancement or text insert with one of those bundled software. Or may have used Photoshop or GIMP to give it more professional look. In any case, the processing techniques once limited to professional circles are now available to us with a few clicks.

In case of grayscale images, the tools of image processing is fairly limited. Gamma-correction is one of the most important image enhancement/restoration/processing techniques. It is based on the logarithmic nature of image display/perception and it rescales the grayscale values of the image for proper display. In hardware-driven image processing, often dealing with analog signals, gamma correction is the most reasonable and effective way to correct image luminance, because of its functional nature of input-output. However, in software-driven, mostly digital signal processing, you can manipulate transfer function in any way you want, not limited to gamma functions.

Here is the darkest image in “There Was A Father”.

There Was A Father, Scene 2538

We all suspect this must be the good example of degraded, badly duplicated, dark frame. It seems inconceivable that Ozu and Atsuta aimed for underexposed look of “Godfather”. So we can enhance the image to give it a new look, right? I tried two different methods of grayscale histogram equalization. First, non-linear grayscale transfer function was applied globally (i.e. whole image). Simple gamma correction was not satisfactory, so the transfer function was adjusted to give a “decent” look. Another method, Contrast-Limited Adaptive Histogram Equalization, CLAHE (1), was also tried. The results are below.

An example of global histogram equalization (Right) and Contrast-Limited Adaptive Histogram Equalization (CLAHE, Left)

First of all, I do not claim these images are something comparable to the works of professional image restoration engineers and archivists. NO. This is only an illustration of what image processing will do to the images. They work on vast pool of knowledge and experience with materials and workings of film and digital images. Here, this is only meant be a demonstration of grayscale transfer function. Second, the methods I utilized are not meant to be ideal. They are far from it. There is no scientifically, technologically sound reasoning for the choice of the transfer functions. CLAHE is probably not suitable at the first place, since it tries to find the best grayscale adjustment locally (i.e. looking at the values at pixels nearby). My point is not how good I can bring this image back. What I want to discuss here is slippery nature of “good” and “original” image.

In the pre-processed image, the left upper corner is almost pitch black, the face of Shuji Sano is barely lit, the details and features sinking into darkness. On the other hand, the hands and papers on the desk are fairly brightly lit, with reflection on the left hand almost saturating. Judging from the image, the major source of illumination is far back in the right, with small spots on desk and background. This is an unbalanced, poorly-lit frame.

The global histogram equalization attempts to bring “back” some brightness in the frame without saturating hands and papers on the desk. It did give some details in Shuji Sano’s face, suits, and the cabinet in the background. Especially, the hands and papers on the desk are quite “lively”. But still half of the face is dim, the left upper corner is dark, and the whole image is still unbalanced.

The CLAHE image gives more push on the shirt and suit, and a bit more balance over the whole area. But it brought unnatural liftup on Sano’s cheek, saturation on some prats of background and paper on the desk.

By then, you might have asked yourself, “What was it like when it was originally released in 1942 Japan?”

An archivist at National Film Center of Tokyo, Fumiaki Itakura (2), discusses there are fundamentally two different approaches to film restoration. One is the approach proposed by Paolo Cherchi Usai, to restore the available image to “model image” or the image at the time of film release. Another is to preserve all the flaws and degradation including scratches and fading as they are. He comments that the digital technologies enable both approaches possible simultaneously.

Here, we have a film, the print of which is poor duplication with many cuts. Duplication was done more than decades ago. The original print might have already degraded even at the time of duplication. Then, restoration with the first approach, the path to “model image”, seems to be a daunting task.

According to another archivist, Fumiko Tsuneishi, at National Film Center of Tokyo, digitization of (color) film prints can vary one laboratory from another (3). Some major processing laboratories already make color adjustment during the transfer. In practice, according to her, this actually provides more flexibility at later stages of restoration. However, as she observes, this throws away the original scan data. She decided that she sticks to the original scan data, even if it might give her a hard time afterwards. In another words, digitization and the philosophy behind it may differ from one source to another. I can imagine some transfers might have done with real-time image adjustment.

The issue here with this film is, looking at this dim image of Shuji Sano, was it meant to be dark? Looking at it, adjusting it, writing scripts to do CLAHE, which is actually suitable for medical images because it brings up slight differences in X-ray images, I realized I would never know. And only thing I know is, I love this image. I love insufficient illumination from far back to throw darker shades. I love other images in this battered print. Many location shootings, with insufficient brightness, uneven grayscales, are much more like what I recollect fondly about landscape in Japan. It may not be “correct”, it may not be as same as the print at the time of release, it may be a “poor transfer DVD” or in my case, TV transfer more than a decade ago, the encounter with the film is personal and individual, not universal. Even if it’s a screening at NFC with condition as ideal as possible, with fellow film lovers, the experience is still personal and individual.

Image processing technology is still evolving, and someday we may have really enjoyable, crisp images of this film. Research and study of the material might bring us the authentic reconstruction of the film, as it was released in 1942. I would love to see it. I really do. But until then, I will appreciate the film as it is now.

As I said, cinema is communication. The original message was transmitted many, many years ago. During the transmission, noise was added, some part of data dropped off, and more noise was added. In digital signal communication, you can actually employ smart signal detection, data error correction, stronger coding for noisy channel, and so on. You can actually reconstruct message perfectly from degraded one. No such fancy stuff here. There is no PRML, erasure code nor ECC. They transmitted it many years ago. I received it. And it is up to me how to interpret it.

(1) MATLAB Documentation,  Contrast-Limited Adaptive Histogram Equalization

(2) “The Present Status of Film Restoration” Symposium 2007 (Japanese, pdf)

(3) “Digital Restoration and Color of Cinema Works” sigCI Symposium, 2005