Colorization

Film colorization (American English; or colorisation (British English), or colorization (Canadian English)) is any process that adds color to black-and-white, sepia, or other monochrome moving-picture images. It may be done as a special effect, to "modernize" black-and-white films, or to restore color films. The first examples date from the early 20th century, but colorization has become common with the advent of digital image processing.

Digital colorization

Computerized colorization began in the 1970s using the technique invented by Wilson Markle. These early attempts at colorization have soft contrast and fairly pale, flat, washed-out color; however, the technology has improved steadily since the 1980s.

To perform digital colorization, a digitized copy of the best black and white film print available is used. With the aid of computer software, technicians associate a range of gray levels to each object and indicate to the computer any movement of the objects within a shot. The software is also capable of sensing variations in the light level from frame-to-frame and correcting it if necessary. The technician selects a color for each object based on common "memory" colors—such as blue sky, white clouds, flesh tones and green grass—and on any information about colors used in the movie. If color publicity stills or props are available to examine, authentic colors may be applied. In the absence of any better information, technicians may choose colors that fit the gray level and are consistent with what a director might have wanted for the scene. The software associates a variation of the basic color with each gray level in the object while keeping intensity levels the same as in the monochrome original. The software then follows each object from frame to frame, applying the same color until the object leaves the frame. As new objects come into the frame, the technician must associate colors to each new object in the same way as described above. This technique was patented in 1991.

In order to colorize a still image, an artist typically begins by dividing the image into regions and then assigning a color to each region. This approach, also known as the segmentation method, is laborious and time-consuming, especially in the absence of fully automatic algorithms to identify fuzzy or complex region boundaries, such as those between a subject's hair and face. Colorization of moving images also requires motion compensation, tracking regions as movement occurs from one frame to the next.

Several companies claim to have produced automatic region-tracking algorithms:

• Legend Films describes their core technology as pattern recognition and background compositing that moves and morphs foreground and background masks from frame to frame. In the process, backgrounds are colorized separately in a single composite frame which functions as a visual database of a cut and includes all offset data on each camera movement. Once the foreground areas are colorized, background masks are applied frame-to-frame.
• Time brush describes a process based on neural net technology that produces saturated and crisp colors with clear lines and no apparent spill-over. The process is cost-effective because it relies on computers rather than human effort, and is equally suitable for low-budget colorization and broadcast-quality or theatrical projection.
• A team at the Hebrew University of Jerusalem's Benin School of Computer Science and Engineering describe their method as an interactive process that does not require precise manual region detection, nor accurate tracking; it is based on the premise that adjacent pixels in space and time that have similar gray levels should also have similar colors.
• At the University of Minnesota, a color propagation method was developed that uses geodesic distance.
• A highly labor-intensive process employed by the UK-based film and video colorization artist Stuart Humphryes, in conjunction with video restoration company SVS Resources, was employed by the BBC in 2013 for the commercial release of two Doctor Who serials: the first episode of The Mind of Evil and newly discovered black and white footage in the director's cut of Terror of the Zygons. For these projects, approximately 7,000 key-frames (approximately every 5th PAL video frame) were fully colorized by hand, without the use of masks, layers or the segmentation method. These were then utilized by SVS Resources to interpolate the color across the intervening surrounding frames using a part computerized/part manual process.
• In 2017, Algorithmia released a free, online colorization service. Although sample limited, the service uses neural nets with artificial intelligence which is trained on colorized images that are fed to deep generative adversarial networks along with automatically generated grayscale.