Restoring a Classic: Rescue from Gilligan's Island

"Rescue from Gilligan's Island," the 1978 made-for-television film, brought the beloved characters of the 1960s sitcom back to life after over a decade. However, as with many films of its era, the movie has faced the test of time, with its film prints suffering severe deterioration, damage, and destruction and the original graphical elements being lost. Moreover, the film has fallen into the public domain, meaning that no one has been willing to put the time and effort into restoring something that anyone can copy and sell. Until now. This article explores the process of restoring this classic film and the technical challenges involved in preserving its legacy for future generations.

The Source Material

There have been a few DVD releases of the film over the years, each offering its own assortment of horrendous quality problems. Unfortunately, although shot on 16mm film like most 1970s television, no film prints are currently known to exist and the only available sources for the movie reside on videotape. The location of the original camera negative is unknown (assuming it even still exists), and prints struck from it have so far been impossible to find. Inquiries to Warner Brothers have gone unanswered. Therefore, we are stuck working from a videotape source, which is far from ideal and creates a whole set of problems of its own. Fortunately, there are different video sources available that were originally made from at least two different film prints, each with its own telltale quirks and characteristics.  That gives us some options when the time comes to patch defects.

One central problem with working from videotape is the frame rate. Film runs at 24 frames per second, while an NTSC standard television displays images at 30 frames per second. To solve this problem, the NTSC video standard uses a process known as a 3:2 pulldown to change the frame rate for broadcast. 3:2 pulldown simply means that each group of four frames is converted into ten separate fields by selectively duplicating fields, which results in a conversion from 24fps to 60fps. Then all the separate fields are interleaved back together to halve the frame rate and create 30 interlaced frames. Because of the particular formula used in the conversion, the resulting output is a pattern of three progressive frames followed by two interlaced frames with combing. That means that 2/5 of the frames will be interlaced. It sounds complicated, but this diagram from Wikipedia demonstrates the process. 

This creates our first problem in that we needed to restore the original progressive 24 fps from this 30fps video. Normally, if we had access to the broadcast master tape we would use a routine process known as inverse telecine (IVTC) to simply reverse the 3:2 pulldown process and leave the original frame rate. However, in this case, all the sources are copies of various generations that are either too poor to accurately detect combing, or the interlaced fields have already been blended (more info about blending can be found here), which cannot be undone by conventional means. Fortunately, there  are still things that can be done. The blended frames generally follow a predictable pattern, so it's possible to selectively "unpick" the blended frames and use AI interpolation to create new, clean frames to replace them and then decimate it back to its original 24 frames per second. It's not a perfect process and requires some manual cleanup, but the result is a relatively clean output that approximates the original frames as closely as we are going to get.

Next, we used digital noise reduction to clean up a lot of the dirt and scratches that were on the surface of the film print when it was scanned. But we did it with a twist, though. We used a technique known as "superscaling" (not to be confused with the Resolve function of the same name) to scale the video all the way up to 8k before denoising. Videotape has inherently low resolution, and so therefore we are "image poor" right out of the gate when working with video. This can't be fixed, but it can be mitigated. When the video is upscaled, new image information is interpolated from the existing data and when the video is downscaled it creates the illusion of sharper resolution and more detail. 

While the video is upscaled, a light pass with DeSpot and RemoveDirtSMC cleans up a lot of the noise without being too destructive on the image. Stronger settings will remove more blemishes but will also remove image detail, so it's generally better to go light and then manually paint out any remaining defects. Once the video is satisfactorily denoised, it's scaled back down to 1080p for the remainder of the work because working in 8k is prohibitively time-consuming. Resolve's DustBuster tool is ideal for manual cleanup, but it requires that each of the film's 134,000 frames be examined one by one - a very tedious and time-consuming process. Using Resolve, our team manually painted out over 10,000 major and minor defects in the frames.

The images below show a sample frame before and after superscaling. The superscaled image looks a little oversmoothed to the naked eye, but in actuality the denoising just revealed what was underneath. It simply exposed that there was never any detail there to begin with. We can hide this to some extent by adding some organic-looking 16mm film grain over the top during the color grade.

Using Resolve and Photoshop, we also cleaned up some severe chroma bleeding at the scene cuts which affected approximately 70% of the cuts in the film. Chroma bleeding causes the color from the last frame in the cut to "bleed over" into an adjacent frame of another scene, and it's a hallmark of videotape sources. Our artists repaired this by either selectively duplicating frames at cuts where there was little or no motion, or by creating new composite frames in Photoshop for scenes with motion where the image stutter of duplicated frames would be more noticeable.

Example of a blended frame

The image showed a lot of instability caused primarily by the gate weave of the film being run through the scanner, as well as some warping that was probably caused by bad splicing of the print. These issues were much more pronounced in the latter part of the film, especially the scenes of the Professor's lab and the Howell dinner party. There were also a handful of instances where the camera made a sudden jerk to the right or left very quickly, and it was difficult to tell whether it was physical camera movement or whether it was introduced sometime during post-production. After some discussion about how true to broadcast we wanted to stay, we decided to remove these movements because they would be distracting to a viewer watching on modern equipment. Even if the movement was there during broadcast, it would probably be seen today as a defect that we failed to fix. After Effects' Warp Stabilizer tool was sufficient to stabilize most of the shots, but a few of the more challenging shots (like Ginger kissing Gilligan at the lagoon) had to be stabilized in Nuke using its superior motion tracking.

Once the primary source was repaired, there were a series of patches that had to be applied from different sources. The Christmas party scene on the Minnow II, for instance, was missing from all of the better quality sources and had to be patched in from an older, lower quality DVD source. The patches all received the same treatment as the main film, and a combination of sharpening and AI upscaling helped mask the quality difference between the sources, although it's still quite noticeable. We were able to match the colors between sources surprisingly well in the color correction process and then give the whole thing a consistent color grade to bring down the black level in the darker areas and make the whites and skin tones look more natural throughout.  We also took the liberty in the color phase of making the sky bluer in several scenes that were clearly overcast on shooting day but were intended to be sunny, as well as bluing up the "ocean" and lagoon water a little bit so it's not so noticeable that it's really just dirty brown water in a tank. 

Lastly, new opening credits were recreated from scratch using parts of the Season 2 DVD release and then using Photoshop to recreate the graphical elements that were unique to Rescue. The opening is identical between the series and the film right up to the point where the beached Minnow zooms in, so we can just use the DVD opening up to that point. From there on, we created a clean plate of the beached Minnow and used Photoshop to recreate graphical elements. Similarly, for the final shot of the island in the opening, we can mask off the island and water and create a clean plate of the sky for a clean title. The ocean water in the movie's original title sequence didn't move because the director used a still image, but we all know that the water was supposed to be animated, so we fixed that. 

The text does not use a proper font that can simply be typed over and is instead a hand-drawn graphic. This is easy to tell because there are alternate versions of the same letter, like the capital "R" below, and upon close inspection, no two letters are exactly identical. They each exhibit some degree of minor variance as a result of being hand-drawn.

That means that each letter has to be treated like any other graphic and manually traced out using the pen tool in Photoshop. This is an incredibly tedious and labor-intensive process that takes a couple of days to complete for all of the opening and post-title credits. We rotoscoped out the credits over the lagoon and created a panorama image of the huts in Photoshop that could be used as a clean plate. That allowed us to fully recreate all the opening titles in a way that's arguably even better than the original.

This special edition will be available on Blu-ray and streaming soon. Check back soon for pre-order information.

Alan Burns May 3, 2023
Share this post