Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
13~32~
SYSTEM AND METHOD FOR COLOR IMAGE ENH~NCEMENT
This application is a division of application Serial
No. 573,741 filed August 3, 1988.
Field of the Invention
The present invention pertains to a system and method for
the color enhancement of monochrome or multi-color images by
relation of digitized gray-scale information from the original
image to unique values (triplets) of hue, luminance, and
saturation derived from a color transfer function applied to
the achromatic gray-scale information.
Backaround of the Invention
In strictly mathematical terms, the image produced on
black and white film is one dimensional (brightness), while an
. original-scene can be viewed as three dimensional (using the
HLS model of color, where hue, luminance, and saturation are
the dimensions). The chemical reaction of the film emulsion,
and the camera mechanisms and filters, essentially reduce the
three dimensional information to one dimension by complex
mapping, and forever obliterate the original hue, luminance,
and saturation information.
Note: It should be pointed out that luminance and
brightness are not the same. Since brightness of the black
and white image results from the complex interactions of the
three elements of the HLS system, there is only a statistical
correlation between brightness and luminance, but they are not
identical.
Color Representstion ~yste~s
Color is both an absolute quality of light (its wave-
length or frequency) and a relative quality which depends on
2 132~32~
the human perceiver. For this reason, various color "gamuts"
have been proposed which allow perceived colors to be
quantified. Among these models is the CIE chromaticity scale,
which permits quantification of color by combinations of so-
called primaries (which are, by themselves, not visible tohuman observers). (See Foley & VanDam, Fundamentals of
Interactive Computer Graphics, Addison Wesley, Reading, MA
198~ pp. 606-608.) In addition to theoretical models of color
(CIE Chromaticity, Hue-Luminance-Saturation, and Red-Green-
Blue), there are many representational systems for colorreproduction in various media such as the RGB system for
cathode ray tubes, the process color system for printing, and
the NTSC, PAL, and SECAM systems for broadcast video. Each of
these models is, in fact, a relatively small subset of both
the theoretical absolute range of colors and of the actual
range of humanly perceivable colors.
One of the theoretical models (Ostwald, W., Colour
Science, Winsor & Winson, London, 1931) which is especially
useful for analysis of color is the HLS model of hue,
luminance, and saturation. This model uses a three
dimensional cylindrical universe (the absolute theoretical
color space), inside of which is a space bounded by a pair of
cones having a common base (the perception space~. The
cylinder's axis (from apex to apex of the cones) is the
luminance axis, with the lower apex having a luminance of zero
(all black) and the apex of the upper cone having a luminance
of 1 (all white). Distance from the luminance axis (normal to
the axis) determines saturation (proportion of a color which
is made up of its complementary color), with the axis
representing zero saturation (equal amounts of the primary
color and its complement and therefore no color) and the
surface of the cone representing maximum perceived saturation
for a given luminance. Finally, angular displacement around
the luminance axis represents hue, as with a conventional
color wheel, with complements separated by 180 degrees.
The solid space defined by the HLS conic model is pointed
at its tcp and bottom to represent the fact that as colors
3 ~324321
become whiter or blacker, they are perceived to lose hue.
Saturation is a characteristic which depends chiefly on the
narrowness or purity of the band of wavelengths comprising a
color. At the exact center of the space is a medium gray
which exhibits a moderate emittance or reflectance for all
wavelengths of light.
The RGB system, is used, for example, in CRT displays,
where each of the three electron quns (red, green, and blue)
is controlled directly by a signal indicating its respective
iO intensity. HLS color may be converted to the RGB
representation (see Foley ~ VanDam, pp. 648-9.), however there
are many HLS combinations which must be "clipped" to be
represented in RGB. The same is true of NTSC video (also
known as YIQ), which has an even more restricted range than
RGB. Thus, not every "real world" color can be reproduced by
video systems using either RGB or NTSC standards. According
to Foley & VanDam, "neither the V in ~SV [Hue, Saturation,
Value] nor the L in HLS correspond to luminance in the YIQ
model, so two different colors defined in either space can
easily have the same luminance, and be indistinguishable on
black and white TV or videotape." ~Foley & VanDam p. 618)
Since approximately 1970, several techniques have been
developed for the addition of color to monochrome images.
Often, these techniques rely on video technology to augment
the video signal in a defined manner, typically by adding a
chrominance component signal to the pre-existing luminance
siqnal. This technique results in an appearance not unlike
that achieved by Thomas Edison in his early efforts to add
tint to black and white film by painting the frames: a
colored area over a black and white picture. Because of their
reliance on video technology, all of these systems suffer from
lack of vividness of colors, and from an inability to
accurately reproduce colors.
,,:
Brief Description of the Invention
~'
The present invention comprises a method for selectively
;
4 132~32~
coloring at least one picture element of an image comprised of
a multiplicity of picture elements by combining image
information ~rom mask-defined regions of the image with color
information, the improvement consisting of: (a) assigning to
an image to be colored, mask-defining region information and
color information from a previous image; (b) moving at least
one of said mask-defined regions a predetermined distance
along a predetermined axis to a moved location and storing
mask-defining region information for said image to be colored
at the moved location.
Brief Description of the Fiaures
Figure 1 depicts in schematic form the creation of two
adjacent frames of graphic information according to the method
of the present invention.
Figure 2 is a schematic representation of a portion of
the image memory structure according to one embodiment of the
method of the present invention.
Figure 3 depicts the replication of a region memory
portion of the image memory structure of the present
invention.
Figure 4 depicts the correction of a value stored in the
replicated region memory portion of Fig. 3.
Figure 5 depicts the application of the HLS color model
to the coloring o~ an array of picture elements stored in the
image memory structure of Figs. 2-4.
Figure 6 depicts a schematic representation of the
"histogram slide" operation according to the method of the
present invention.
Figure 7 depicts a schematic representation of the
"histogram multiplication down" operation according to the
method of the present invention.
Figure 8 depicts a ~chematic representation of the
"histogram multiplication up" operation according to the
method of the present invention.
Figure 9 depicts a schematic representation of the "color
132~32~
bending" operation according to the method of the present
invention.
Detailed Description of the Invention
General Process overview
The film to be colored or color enhanced is first
digitally recorded at a preselected resolution. In instances
where the final product will be placed on conventional video
tape, appropriate resolutions are employed. (E.g. NTSC=525
lines, PAL=640, HDTV=1125-1500). Where the final product will
be used for theatrical presentation, resolutions of 1024 lines
or more may be employed. The digitization process (depicted
in Fig. 1) compri6es projecting each frame of the original
(typically black and white) film through a liquid gate device
which masks scratches and other imperfections. The projected
; 15 frame is captured using a video camera such as a saticon tube
and the video signal is digitized and stored in a random
access memory frame buffer. The digitization process allows
ad~ustment of the intensity of light falling on the film in
order to achieve the best contrast for the particular frame
being digitized (gamma correction). After acquisition of a
frame, the digital representation o~ the frame is stored in
non-volatile storage for ~urther processing. The digitization
process is repeated on a frame-by-frame basis, marking each
frame with a time and sequence indication (SMPTE time code).
For each scene (sequence of frames), a frame is selected
as being representative of the majority of the elements
present in the sequence. This "key frame" is used as a
template for the enhancement o~ other frames within the scene
sequence. For this reason, the greatest care is applied to
the processing of key frames.
After kay frames are color-enhanced and defined, they are
passed with the rest of the frames comprising a scene to a
work station where the regions defining the application of
6 1~2~32~
color are tracked from frame to frame. After each frame has
been enhanced, it is restored in non-volatile storage and sent
to a central storage where all enhanced scenes are eventually
reassembled according to their original sequence as a final
product motion pictuxe.
Key Frame Coloring
The first task of the key frame colorist is to identify
areas of the image which comprise objects which will be
designated to be of similar hue. Several software tools are
available to the colorist to aid in this computer-interactive
definition process.
A "free hand" mode allows drawing and erasing as well as
control over a variably-sized cursor. A pointing device
controls an interactive cursor displayed on a screen
overlaying the key frame to be processed. Movement of the
pointing device effects an analogous movement of the cursor
and the picture elements selected by this movement (which are
displayed on a graphics plane known as the "mask plane") are
used to define the edge of an area. Because the cursor size
ma~ be reduced to only one pixel, the smallest region which
may be produced is also one pixel.
A second drawing tool available to the colorist is one
which defines polygons, rather than completely free-hand
shapes. Using the pointing devices, the colorist may indicate
the precise placement of vertices which are intended to define
the perimeter of a polygonal area. Each vertex may be
assigned or removed and, when complete, the vertices may be
completed to define a closed path.
After definition of a closed perimeter by free-hand or
polygon methods, the enclosed area may be assigned to the
region (by setting the bits in the mask plane) by using one of
two filling algorith~s. An "area fill" assigns all picture
elements within the closed shape and up to the defined edge to
; the area. A "cartoon fill" combines the defined edges with
edge recognition algorithms in order to fill an area which is
7 132~321
bounded by a border line such as the black pen lines which are
typical of cartoons. Cartoon fill mode fills up to, or
slightly over such heavy border lines.
After areas have been defined for all objects having a
common hue, the final step is the selection and application of
a given hue to that area. ~olors may be selected by
interactively pointing to a color wheel displayed on a video
screen, and then by manipulating control indicators on the
video screen which allow refinement of the gray-scale
information for the area as described below. After selection,
the color may be displayed by its application to the areas
(which are collectively known as a color mask.) After a color
mask has been defined, its associated picture elements are
identified as a "region", the mask plane is cleared, and
additional masks are defined in a similar manner, one for each
hue which will be displayed in the frame.
Once all regions are defined, the regions, the underlying
frame, and the colors assigned to each region are stored for
later retrieval by colorists who will reiterate the process
using the subsequent frames of each scene and the key frame
which has been defined.
An additional tool for defining regions is the threshold
region definition tool. The threshold region definition tool
permits the automatic definition of regions to be colored.
First, a brush size (a form of cursor) is selected. Next, an
upper or lower threshold gray-scale value is selected. The
upper/lower gray-scale value is representative of the edge of
a con~iguous group of picture elements having common
brightness. An upper gray-scale value is selected if the
threshold region definition tool is to be used to select
picture elements below a given upper threshold value.
Alternatively, a lower threshold gray-scale value is selected
to designate a region above said lower threshold value. The
picture elements located within the brush tip are analyzed to
determine the medium brightness o~ those picture elements. If
the medium brightness within the brush tip is within the
designated threshold region (either below the upper threshold
8 132~321
region or above the lower threshold region), the picture
elements are included in the designated region. The picture
elements above an upper threshold region or below a lower
threshold region are excluded from the new region.
The process of extrapolating color enhancement from key
frames to entire scenes is performed by identifying and
tracking motion as well as appearances and disappearances of
objects. Generally fewer than 5% of the picture elements in a
frame of a scene change in the succeeding frame. ~ecause
motion pictures are filmed at 24 frames per second, increments
of motion on all but the most rapidly moving objects are
extremely small and easily followed. Once motion has been
identified, adjustment of the regions and application of the
key frames' predefined colors to those regions is all that is
required.
Implicit in all frame-to-frame tracking is the ability to
copy predefined parts of the preceding or key frame's re~ion
memory contents and to use those parts in subsequent frame
processing. After copying, small adjustments may be made to
the new region memory contents and they may be saved with the
new frame. Importantly, because the digitization process
assures consistency of brightness and gray scale, and because
colors are predefined by an art director working with ~ey
frames, overall consistency ~rom frame-to-frame, and scene-to-
scene is also assured.
The simplest form of tracking from frame to frame is
where motion in the frame is apparent only as a result of a
camera pan. In such instances, the entire region memory may
be copied from that of the preceding frame and "shifted" in
the direction of the pan. Such a shift will create an
undefined area at the end opposite the motion. This area must
then be added to the existing regions in order to completely
apply color.
When only some of the objects of a frame move, a
corresponding portion of the region memory may be translated
as well. In this operation, that portion of the region memory
which is to be moved i8 designated by the operator and is
9 1~24~21
moved manually to its new location, thus overwriting the
former contents of the region memory at the new location. The
old location must then be re-defined and/or added to one or
more pre-existing regions in order to apply color.
In highly complex scenes, where several types of motion
are present, (especially compound motion of subject and
camera) a "differencing" procedure is employed. A frame for
which areas have been de~ined and to which color has been
applied is subtracted from a next subsequent frame in order to
reveal those picture elements which have changed (in gray-
scale value). These areas of difference are then displayed as
a contrasting overlay on the subsequent frame to be processed.
Alternatively, an algorithm for edge detection in gray-scale
images may be used and may be correlated to the previous
frame's region ~emory. Detected differences may then cause
the subsequent frame's regions to be adjusted or deleted, as
required. By examining the coincidence of this overlay and
the region structure whi~h has been duplicated from the
preceding frame, the production colorist is able to quickly
designate areas which should be added to the region structure,
and those which must be removed from a given region and
reassigned.
The final operation is to assure that no objects have
either completely appeared or disappeared from the frame. If
an object disappears, its region must be removed. Similarly,
newly appearing objects must have a region defined for them
"on the fly" and a color assigned as well. After adjusting
the region for the next subsequent image, the colorist again
stores that image to~ether with its region memory and repeats
the process for each subseguent Prame.
To permit quality control and review of a sequence of
color-enhanced frames, two modes of viewing are used: Cine
mode and Cycle mode. In cine mode, a series of frames is
sequentially displayed, shuttling first in a forward
direction, and then immediately in reverse. The speed may be
controlled by the operator. In cycle mode, display direction
and speed are directly tied to the pointing device to allow
10 132~321
fine control over viewing frame-to-~rame transitions.
Hardw~re Overview
The method of the present invention is implemented on a
special purpose version of an RTI-Station image processing
system interfaced to a personal computer (such as the IBM*
personal computer AT). The RTI image processing subsystem
(available from Recognition Technology, Inc. of Holliston, MA)
comprises a video digitizer, a pipeline architecture pixel
processor, and image memory which is all interconnected both
by a host computer bus and by a synchronous video bus. The
video digitizer takes its input from a video camera such as
that described above for digitizing film frames, and outputs
its information to a high definition color CRT (though not
necessarily one interfaced by conventional NTSC video
circuitry). The host computer bus permits the image
processing subsystem to communicate with the user through the
host CPU, its memory, associated mass storage devices, and I/O
devices including pointing devices and keyboards.
The image processing subsystem of the preferred
embodiment of the present invention comprises an image memory
having 51Z x 512 spatial resolution, 8 bits of brightness
resolution, and 4 bits for region identification.
Alternatively, the same system may be configured for 6 bits of
brightness resolution and 6 bits of region identification or
for 7 bits of brightness resolution and 5 bits of region
identification. (One additional bit is reserved for use by
the system in construction of the mask bit plane.) General
information regarding the structure and operation of the RTI
Station and its associated software library (RTILIB) may be
found in the RTILIB/500 Tutorial Revision l.Oo and the
RTILIB/500 User Manual Revision A, which are published by
Recognition Technology, Inc.
Although ~he currently used implementation employs a 512
* Trademark
11 ~324321
line spatial resolution, it will be recognized by those
skilled in the art that increases to and above 2000 line
spatial resolution are within the scope of the present
invention. Additionally, the image memory may be a part of a
larger frame store memory which may be, for instance, 4096 x
4096 x 16 bits. The display of frames may then be controlled
by a memory pointer into this frame store.
For the purpose of color selection, the host CPU is
provided with a 24 bit RGB video board (such as the Targa*24,
a product of AT~T Information Systems, Inc.) A color wheel
which corresponds to the common base of the HLS cone is
displayed on a video screen. Using a pointing device, the
colorist selects a particular color on the wheel, thus
defining both hue (angular position on the color wheel) and
saturation (distance from the center). As deRcribed more
fully below, this unique 24 bit RGB value is translated
immediately into the HLS coordinate space, correlated to the
modified image gray-scale, and retranslated for output on a
RGB monitor.
~Ortware overvlew
Referring now to Figs. 1-4, according to the method of
the present invention, a series of digitized monochrome images
may be stored in a random access memory of a computer as an
array of gray-scale values. For example, 8 bits of
information may be used to represent each picture element,
thus permitting a total of 256 unique gray-scale values to be
represented. (Of course, more bits may be used to achieve
better resolution and dynamic range for the gray-scale
` information.) The digitization step may optionally include
the use of look-up tables in order to enhance contrast, adjust
gamma, enhance edge definition or "stretch" the gray-scale of
a faded image. Associated with the memory used to store the
gray-scale values is a region memory used to uniguely identi~y
* Trademark
1324321
each picture element with one of a number of regions (depicted
as a four bit memory structure in Figs. 2-5). In addition to
the image memory and region memory structures, a mask plane
memory is also associated with the image for use in
identifying picture elements. T~rough conventional
techniques, the contents or parts of the contents of these
memory structures may be copied from one structure to another
(for example, from the region memory to the mask plane memory)
or from one frame's memory to another (as depicted in Fig. 3
for the transfer of frame 1 region memory to frame 2 region
memory.) The copied contents may, of course, be altered by
the operator to conform to the underlying frame contents as
depicted in Fig. 4.
As depicted in Figs. 5-9, the colorist has at his or her
disposal a number of software-implemented tools which control
the construction of the color transfer function. For
computational efficiency, the luminance axis of HLS
cylindrical space is represented as an integer range of
[0..1000] (as opposed to the conventional representation as a
20 real number in the range [01.0]). The gray-scale of a frame
is linearly related to the integer range by defaulting a gray-
scale value of 128 to a luminance value of 500, and
corresponding each discrete gray-scale unit to each integer of
luminance. (Thus, 255 gray-scale corresponds by default to a
luminance of 627.)
In order to affect the color transfer function
constructed by the system, the colorist actually manipulates
the frame gray-scale information by applying histogramming
techniques such as translation, compression, and stretching.
Specifically, the entire gray-scale may be made to appear
either more or less luminous by a simple "slide" of the gray-
scale (GS) with respect to the luminance scale (Fig. 6).
Instead of corresponding GS - 128 to L = 500, the operator may
instead choose to brighten the frame by choosing L = 750.
Similarly, the frame may be darkened by choosing L = 350. As
depicted in Fig. 6, the portion of the color transfer function
falling ~ithin the HLS conic space which corresponds to the
13 132~321
gray-scale is controlled by this slide operation. At its
extremes, the operation creates a threshold in the transfer
function beyond which all gray-scale values are displayed as
either white (above the upper threshold), or as black (below
the lower one).
void slide map(); /*slide brightness mapping */
void stretch map(); /*stretch or compress brightness mapping ~/
void bend on(~; /J'enable color bending */
void bend off(); /~disable color bending ~/
void slide bend(); /*slide region of color bending */
void stretch bend(); /~stretch region of color bending */
/*LOCAL CONSTANTS--------- */
#define TRUE 1
#define FALSE 0
/*Local Static Data---------------~/
static int brt I = 0; /*brightness value for start of ramp ~/
static int brt u = 1000; /*brightness value for end of ramp */
static int bend = FALSE; /~logical - TRUE if color bend enabled */
static int markl; /*lower color bending overlap marker ~/
: static int mark2; /~upper color bending overlap marker ~/
/~Global Data (and externals)----*/
extern int hue 1; /*primary hue value - [0.. 360] */
externintsatl; /~primarysaturationvalue-[0.. 1000]~/
extern int hue 2; /~secondaryhuevalue - [0.. 360] ~/
extern int sat 2; /~secondary saturation value - [0.. 1000] ~/
/~secondaries valid for bend = TRUE only ~/
void slide map (dbrt)
: int dbrt; /~change in brightness - [-999--9991 ~/
{
/~make sure that the requested change is in range */
dbrt = max (dbrt, -999);
dbrt = min (dbrt, 999);
- /~check for limiting case ~/
if ( (dbrt ~ 0) && (brt I < -dbrt)) {
stsetch_map (-brt 1,1000-brt u);
return;
}
/~Simple cæe: appiy the change ~/
brt I += dbrt;
brt_u ~ = dbrt;
if( bend) {
markl += dbrt;
mark2 += dbrt;
. }
~/ End Slide map ~/
14 13243~
Another operation available to the colorist is a
ratiometric multiplication of the gray-scale by a value
(depicted in Figs. 7-8). When the multiplier value is a
fraction less than one, a ratiometric compression of the gray-
scale occurs which relatss the 256 discrete gray-scale values
to fewer than 256 integer luminance values. When the
multiplier value is greater than unity, a "stretch" of the
gray-scale occurs which results in a correspondence of the 256
gray-scale values to a range of luminance greater than 256.
These multiplications may also occur using either end of the
gray-scale as a fixed point, yielding 4 variations as depicted
in Figs. 7-8.
Listing 2
void stretch map( db low, db up)
int db low; /~ change in lower ramp point ~/
int db up; /~ change in upper ramp point ~/
int newbrt 1, newbrt_u, newrnark_1, newmark_u;
double fmarkl, fmark2, rlen, newrlen;
2 0 t~ Check that change parameters are within range ~/
db low ~ min( db low, 999);
db low = max( db low, -999);
db up = min( db_up, 999);
db up = max( db up, -999);
2 5 /~ Apply change to lower ramp point ~/
newbrt_1 = brt 1 + db low;
newbrt 1 = max( newbrt 1, 0);
newbrt 1 = min( newbrt_l, 99~);
/~ Apply change to upper ramp point ~/
newbrt u = brt_u + db_up;
newbrt u = max( newbrt u, 1);
newbrt u = min( newbrt u, 1000);
/~ Make sure that the new combination is valid ~/
if( (db_up == O) && (newbrt 1 >= newbrt u) )
newbrt_1 = newbrt u- 1;
else if( newbrt_u ~ = newbrt_1)
newbrt_u = newbrt 1 + 1;
/~ Reset color-bend markers for new ramp ~/
if( bend){
4 o rlen = (double)( brt_u - brt_1); /~ current ramplen ~/
newrlen = (double)( newbrt_u - newbrt_1); l~new ramplen ~/
fmarkl = (double)( markl - brt 1) / rlen;
fmark2 = (double)( mark2 - brt ~) / rlen;
132~32
markl = (int)( f~narkl * new}len) + newbrt 1;
mark2 = (int)( ~nark2 * newrlen) + newbrt 1;
/~ Move new ramp points into official ramp points ~/
brt 1 = newbrt_1;
brt u - newbrt u;
} /~End of stretch-map ~/
In depicting reflected light (either originating from a
colored source, reflecting from a colored surface, or both),
it is necessary to ~bend~ the color transfer function. As
shown in Fig. 9, the colorist has available the option of
setting a pair of "bending points" for the function which
determine the nature and extent of the bending. These points,
depicted in Fig. 9, (as points A and B) define the endpoints
of three line segments (OB, BA, and Al). Of these segments,
the first and last each have associated with them unique,
operator-selected values of hue, while the AB segment
represents a mixture of hues and saturations. The locations
of points A and B are defaulted to 25% and 75% of the current
gray-scale range, thus providing color mixing over one-half of
the gray-scale. These locations, however, may be changed by
the operator to change the "rate" of color bending.
Listing 3
void bend_on()
{
int midpoint, ramplen;
if( bend) return; t~ already done, do nothing ~/
ramplen = (brt u - brt 1 + 1~ length of current ramp ~/
midpoint = brt_1 + ramplen n; /~ midpoint of current ramp ~/
3 0 markl = midpoint - ramplen /4; /Y bottom of color overlap ~/
mark2 = midpoint + ramplen /4;1~ top of color overlap ~/
} /~ End of bend_on
void bend off()
bend = FAISE;
} /~End of bend-off ~/
void slide bend( dbend)
int dbend- /~ change in region of bend - -999 to 999~/
16 l32432
int ovlplen;
/~ If bend not enabled, do nothing */
if( Ibend) return;
/~ Make sure that there is room to go in the requested direction ~/
if~ ~ dbend < 0) && (markl == brt 1) ) return;
if( (dbend > 0) && (mark2 == brt u) ) return;
/* Make sure that the requested change is within raDge */
dbend = (dbend < 0) ? max ( dbend, brt 1- markl):
n~in( dbend, brt u - mark2);
/~ Apply the change ~/
markl += dbend; /~ new markl position ~/
mark2 += dbend; /~ new mark2 position ~/
} /* End of s1ide bend ~/
void stretch bend( sfact)
int sfact; /~ stretch factor (-100 to 100)~/
{
double ~act, hmarklen, markfact, marklen, midpoint, ramplen;
/~ Compute a floating point stretch factor ~/
fact = (sfact > = 0) ? 1.0 + (double)sfact / 100.0;
1.0 - (double) (-sfact)/100.0;
/~ Get floating point ramp, overlap length and midpoint of bend */ ramplen = (double) brt u
- brt 1 + 1);
; marklen = ldouble) ( mark2 - markl + 1);
midpoint = (double)markl + marklen/2.0;
2 5/~ Compute and adjust the percentage overlap ~/
~ markfact = marklen I ramplen;
: markfact ~- fact: /~ apply change ~1
- hmarklen = (ramplen ~ markfact) / æo; /~ half new mark length ~/
/~ Compute ~ew mark points ~/
30 markl = max( (int)( m~dpoint - hmarklen), brt_1);
~ mark2 = min( (int)( midpoint + hmarklen), brt_u);
} /~ End of stretch_bend ~/
As an example of bending, consider the selection of an OB
hue as flesh tone, and the Al hue as orange (as would be the
case if a face were viewed by reflected fire light). The BA
segment would contain hues which are flesh tones in the less
bright areas, but for brighter ones, take on more and more
orange appearance. Of course, it will be appreciated that the
selection of the locations of points A and B or the inclusion
of additional bending points are fully within the spirit of
the technique.
Because regions are defined so as to include all objects
of a similar hue, regions contain a wide range of picture
element gray-scale values. In order to accurately represent
the gray-scale values in the final colored image, a color
. . .
.~.- .
.,, i~ ,
17 ~32432~
transfer function must be define~ for converting the one-
dimensional gray-scale to a set of three dimensional HLS
values. From the point on the displayed color wheel chosen by
a key frame colorist, a line segment is extended vertically
parallel to the a~is of the cylindrical HLS universe. This
line segment passes through the points in the HLS cones which
represent the colors to be applied within the region, and
comprises the color transfer function. For example, an 8 bit
picture element having a gray-scale value of 190 is
represented on the gray-scale of Figure 5. The corresponding
point on the upper segment of the color transfer function
defines a unique HLS value which will be applied to those
picture elements within the region having the 190 gray-scale
value. Where the line segment does not fall within the HLS
conic space, the representation is either pure white (in the
region above L = 0.5) or pure black (in the region below
L = 0.5). It will be recognized that because the color
transfer function traverses a portion of the altitude of HLS
coordinate space, points which fall at either extreme of the
gray-scale are correspondingly less saturated when displayed
in color. This renders the displayed colors much more
realistic than systems using constant saturations.
Specular highlights are those areas of particularly
bright reflection due to angle of incidence effects. The
production of colorless specular highlights can be produced by
applying the histogram stretch technique described above to
"pull" brighter gray-scale values outside of the HLS cone,
thus making them appear white. The extent of this stretch
controls the appearance of the specular highlight.
Referring again to Figure 9, there is depicted a
schematic representation of the method by which both specular
highlights and reflection of colored light sources may be
achieved. Although human perception is, in fact, best
represented by the HLS double cone, the actual physics of
color and light are best represented by an HLS cylinder which
has luminance as its axis, and saturation as its radius.
Within this cylinder is inscribed the conic perceptual space.
18 13 2 43 21
This conic space defines the limits of human perception, and
the preGise shape and size of the cones varies from person to
person. (A colorblind person, for example, has a cone with a
"flattened" side, while one with night-blindness has an
extremely truncated lo~er cone.) By manipulating gray-scale
information representing an image, the color transfer function
by which gray-scale is transformed into hue, luminance, and
saturation is affected, and a variety of effects are produced.
Glo~sary o~ Terms
Image - monochrome or color analog picture
Picture Elements - discrete locations in an image
Brightness - the absolute amount of light transmitted through
a monochrome image (dependent on incident light
intensity)
Capturing an Image Digitally - recording digital data
representative of an image and from which an image can be
re-created.
Gray-Scale Value - relative amount of light transmitted
through a point on a monochrome image (dependent on
incident light intensity)
Luminance - the absolute amount of light emitted by, or
transmitted through a point on a color image
Hue - the peak or primary wavelength emitted by or transmitted
through a point on a color image
Saturation - the relative number of primary to complementary
points in a given area of a color image (more primary =
higher saturation)
Region - a data structure in memory used to identify a group
of picture elements to which a defined color transfer
function is to be applied.
Mask - a data structure (commonly a graphics plane) in memory
used to identify a group of picture elements to be acted
upon, such as ~y their inclusion within a region.
Map - a data structure (commonly a multi-bit graphics plane)
for identifying one or more regions of an image, each such
region comprising picture elements representative of one
or more objects or image areas having the common property.
Specular Highlight - region of high brightness (monochrome)
or high luminance (color) brought about by direct (angle
of incidence) reflection of a light source.
19 1324~2~
Statement of Industrial Utility
The method and apparatus of the present invention is
use~ul in the color enhancement of achromatic images such as
still photographs, motion pictures, video taped images,
telemetry data, and the like.
