―316― J. Soc. Photogr. Sci. Technol. Japan, Vol.64, No.

5, 2001

技 術 レポー ト Image Processing of Digital

Minilab System

デ ィ ジ タ ル ミニ ラ ボ の 画 像 処 理

Fumihiro SONODA*, Tohru MATAMA* and Hiroaki NAKAMURA*

園 田 文 博*・ 真 玉 徹*・ 中 村 博 明*

Abstract Digitalization has been progressed on Minilab systems in recent years. The advantages of
digital Minilab are not only printing from negatives alone, they can also print from a various input
sources including reversal films and DSC, with greatly improved quality through computation of digital
image processing and a wide variety of processing. However, the computation cost for image process-
ing is extremely high, and software processing does not ensure productivity. To solve this problem, we
have succeeded in realizing the algorithm with hardware. The present report describes this method.

要 旨 近 年 ミニ ラ ボ の デ ィジ タ ル化 が 進 んで い る。 デ ィジ タ ル ミニ ラ ボ の優 位 性 は ネ ガ だ け で な くリ
バ ーサ ル やDSCな ど多 様 な入力 ソー ス か らの プ リン ト,デ ィ ジタ ル画 像 処 理演 算 に よ る画 質 向 上 や画 像 の
様 々 な加 工 に あ る。一 方,画 像 処 理 の演 算 コス トは非 常 に高 く,ソ フ トウ エ ア処 理 で は,生 産 性 が確 保 で き
な い。そ こで我 々 はアル ゴ リズム をハ ー ドウエ ア化 す る事 で この 問題 を解 決 した。本 レポ ー トで は この実 現
方法 につ い て言 及 す る。

Key words: Minilab, image processing, hardware

キ ー ワ ー ド:ミ ニ ラ ボ,画 像 処 理,ハ ー ドウエ ア

if complicated computation is performed to get

1. Introduction
high quality. So solving this problem is
Digitalization of Minilab systems has been extremely important.
progressed in order to cope with diverse input In an effort to find a solution to this problem,
sources including digital cameras and a wide we used the hardware for the image-processing
variety of printing method including super- algorithm instead of using the software, thereby
imposition of images, since difficulty provide ensuring compatibility between high image qual-
high quality image processing by previous ana- ity and high degree of capability.
log method. In the meantime, due to the The present report introduces an implementa-
Minilab system is a production machine for tion of the image-processing algorithm, mainly
photo print, productivity and profitability are configuration of digital Minilab system "Fron-
important factors. The Minilab system is tier".
required to increase productivity and reduce the
2. System configuration of the digital Minilab
costs.
system "Frontier"
Image processing requires a huge amount of
image data in order to achieve higher quality The digital Minilab system comprises <1> a
than that of conventional printing. As a sub- film scanner where the film is photographed by
stantial reduction of product capacity will result CCD and is converted into digital data, <2> an

Received 12th, July 2001, Accepted 4th, September 2001 平 成13年7月12日 受 付,平 成13年9月4日 改 訂 受 付 ・受 理
*
Fuji Photo Film Co., Ltd. Miyanodai Technology Development Center, 798 Miyanodai, Kaisei-machi, Ashigara Kami-Gun,
Kanagawa Pref. 258-8538
富 士写真 フイルム株式会社 宮台技術開発 センター 〒258-8538神 奈 川県足柄 上郡開成町宮台798
 ―317―
Vol.64, No.5, 2001 Image Processing of Digital Minilab System

image processor where images are processed to

suitable one for printing, <3> an exposure unit to
write the image on photo paper by laser, <4> a
developer to develop exposed paper, and <5> an
operation unit for determining image processing
and printing conditions.
The print data from the digital camera or
other image files are sent to the image processor
via the operation unit <5> (Fig. 1).
The system sequence can be broadly classified
into three steps; (a) a pre-scanning step for
obtain an image to determine the scanning and
image processing conditions, (b) an inspection
Fig. 2 Frontier Operation Sequence
step for checking and correcting print, and (c) a
fine scanning step where a printed image is creat-
ed according to the result of inspection. In the 3. Implementation of image processing algor-
pre-scanning sequence, one complete film is scan- ithm
ned continuously under fixed scanning condition,
thereby obtaining images for all frames. In the Being a print product machine, the Minilab

setup sequence, as a result of computation is system has to be capable of printing 1,000 to

2,000 sheets per hour. The pre-scanning
performed based on this all film's data, setup
sequence is taken into account, image process
performance is improved. Scanning conditions
for fine scanning and various parameters for computation must be completed about one sec-

image processing are determined by this compu- ond per frame approximately. Moreover, the

tation. In the inspection sequence, this image is number of pixels exceeds 6 megabytes in terms

displayed on the screen and the operator is re- of 300 DPI in 4R-size (102 •~ 152 mm) printing, so
image processing must be performed in 500
quested to confirm the image or correct it if
required. The result of this manual key correc- nanoseconds/pixel or less.
Since we considered it difficult to use software
tion is reflected in the fine scanning conditions
and image on the screen. In the fine scanning to perform the computation (discussed later)
within this time limit, we began to study the use
sequence, scanning and image processing are
carried out according to the image processing of hardware. However, we use software to
obtain the image processing parameters for the
parameters determined by computation from
setup computation from the following reasons:
setup and inspection. After image processing,
the data is sent to the exposure unit where it is <1> the computation is processed for not real

exposed on paper by means of a laser. The print images, and <2> this processing is not
formalized, hardware is unsuitable for this pur-
exposed paper is fed to the developer where it is
developed. This sequence of operations is per- pose.

formed by pipeline processing, thereby ensuring 3.1 Basic configuration of image processing
hardware
a high degree of productivity (Fig. 2).
Many ways of using hardware can be consid-
ered. Our "Frontier" utilizes a method called

pipeline processing. In this method, an image

processing circuit is arranged on the image
transfer channel. When an image is transfer-

red, image processing is carried out automati-

cally. Although there is a time delay from input

to output, productivity does not fall. This

method may not satisfy to obtain a print that

want immediately, but it is ideal for continuous

Fig. 1 Frontier System Configuration processing of many prints in a specified time as

―318― Fumihiro SONODA et al J. Soc. Photogr. Sci. Technol. Japan

in the case of Minilab system (Fig. 3).

With the Frontier, this pipeline system has
been used to configure the entire image-
processing unit and provide computation for
image processing. The image processing
parameters are determined by setup computa-
tion and pre-set in to the hardware. According
to the operation sequence, process is the perfor-
med automatically merely sending images (Fig.
4).
The other characterized technique of Frontier
Fig. 5 Two-dimensional Processing Converted into One-
image processing is conversion of two-
dimensional Processing
dimensional processing into one-dimensional
processing. According to this method, the han-
dling as in filtering process for two-dimensional corrections, high-speed memory is provided for
space is divided into two-direction processes the LUT (Look-Up Table), and the parameters
(horizontal and vertical) and performed twice in are set in this memory by software previously .
one-dimensional processing. This method Looking up is made to this table in real time
results in downscaling of hardware. As for the during processing. In addition, the time-sharing
hardware configuration, an image memory is method is used to reduce the number of large-
installed between the two one-dimensional scale circuit blocks including the multiplier.
processing circuits. After the image processed This method is used throughout the system .
in the horizontal direction has been written in Even so, the size of the circuit boards will be
the memory, it is read out after 90-degree rota- huge and system feasibility cannot be realized if
tion, and vertical processing is performed (Fig. commercially available ICs alone are used. To
5). solve this problem, ASICs (application specific
Several table used for computation of various integrated circuit) are used throughout the
image processing hardware.
3.2 Implementation of various types of
image processing
3.2.1 Correction of distortion aberration
and lateral chromatic aberration of
the lens
The Frontier has a built-in function for cor-
recting the distortion aberration and lateral
chromatic aberration of less costly plastic lenses
having inferior lens performance (Fig . 6). In
Fig. 3 Pipeline Processing the practically, this is effective to the Image
from a camera which can presume the lens char-
acteristic like single use camera .
For G (green) in this processing, the amount
of distortion aberration is calculated from the
approximate lens data, and according to this
amount of distortion aberration , Zoom
magnification is changed for every coordinates
at the time of zoom in/out processing .
For R (red) and B (blue), the deviation from
G is calculated from the approximate lens data ,
and zoom in/out processing is carried out as in
the case of G. Downscaling of hardware is
Fig. 4 Frontier Image Processing Block Diagram achieved by combination of zoom in/out and
 ―319―
Vol.64, No.5, 2001 Image Processing of Digital Minilab System

Fig. 6 Conceptual Diagram for Correction of Distortion

Aberration and Lateral Chromatic Aberration

Fig. 8 Hardware Configuration

(Distortion Aberration and Lateral Chromatic
Aberration Correction, and Zoom In/Out Process-
ing)

weighting factor. Since this mask image is a

low frequency image, it is stored in the memory
in compressed form. It is decompressed on a
real-time basis when correction is made. After
being multiplied by the weighting factor, it is
Fig. 7 Distortion Aberration and Lateral Chromatic added to the original image (Fig. 9).
Aberration Correction Algorithm This processing is provided by an ASIC, a
memory for storing the peripheral darkening
aberration correction processing as already dis- mask and SRAMs (static random access mem-
cussed (Fig. 7). ory) for the LUT.
Originally, lens data used for correction is 3.2.3 Improvement of facial expression
expressed in two-dimensional higher order The major portion of the subject is a human
terms. By using a one-dimensional higher order face in many cases. The print yield can be
expression to approximate this, we created an improved by recognizing the face and by correct-
ASIC for one-dimensional processing. Further- ing the density based on face density. This is
more, this higher order expression originally especially effective for scenes (e.g. shooting in
required floating-point operations to be perfor- flash mode) , which have not been handled suc-
med. To ensure downscaling of hardware, how-
ever, only fixed point operations are performed
based on the approximation method. In case of
Frontier, this processing is performed by two
ASICs (a line memory for timing adjustment, an
image memory for rotation and image memory
control circuit) (Fig. 8).
3.2.2 Correction of peripheral illumination
level
A less costly lens has a greatly reduced periph-
eral illumination level in addition to distortion
aberration and lateral chromatic aberration.
Fig. 9 Correction of peripheral illumination
Consequently, a function for correcting the Correction of peripheral illumination is performed
peripheral illumination level is also provided. using the mask image data of the same position as
In this processing, a mask image for reversing a scanning image.
dimming correction is created from the approxi- Since it is reduced, a mask image also performs
mate lens data after being multiplied by a expansion processing on real time.
―320― Fumihiro SONODA et al J. Soc. Photogi. Sci. Technol. Japan

cessfully by the conventional algorithm. The
image obtained in the pre-scanning process is
used to extract the face. We have improved
accuracy by using a combination of two extrac-
tion methods, extraction by color and shape, and
extraction by profile. In the case of Frontier,
the result of face extraction is also used for local
tone control of the face in addition to simple
density correction.
To avoid unnatural connection between the
extracted face area and other areas, only the
ultra-low frequency area is extracted to gener-
ate a facemask. This is weighted to control the
density of the face in the fine scan image.
To perform this processing on a real-time
basis, the facemask is preset in the parameter
memory for two-dimensional correction in the
hardware. When the image is transferred, this
data is read out sequentially and correction is
provided. To reduce the cost of this memory
and setup time, the facemask is written in com-
pressed form. When it is read out, it is decom-
pressed on a two-dimensional basis (Fig. 10) .
This processing is provided by an ASIC, a
memory for storing the facemask and SRAMs
for the LUT.
3.2.4 Processing of Hyper-Sharpness
To emphasize sharpness, USM (un-sharp mas-
king) is generally used. This processing
involves emphasizing the grain of the film if
degree of emphasis is increased. To solve this
problem, we used the characteristics that the
grain of the film has high random nature and
color correlation in low.
We have separated the spatial frequency zone
of the image into three portions. If color corre-
lation in the medium/high frequency area is
high, will be the edge of the image and increase
the gain in the high frequency area. If it is low,
it will be flat part in picture and reduced the gain
in the medium frequency area (Fig. 11) .
The medium/high frequency components are
converted into the luminance component by the
Adaptive Matrix (matrix circuit for making the
luminance component generation coefficient var-
iable according to the pixel value) , and computa-
tion is performed (Fig. 12) .
To downscale the hardware, this processing is
also converted into one-dimensional processing,
and is performed twice in the horizontal and
vertical directions after rotation in the image
memory.
3.2.5 Processing of Hyper-Tone
When a scene with wide dynamic range has
been photographed, it is difficult to represent
both the light and dark portions on prints simul-
taneously. We have solved this problem by
implementing digital dodging (auto mask
effect) . In Hyper-Tone processing, the ultra-
low frequency component of image brightness is
extracted, and this value is used to compress the
dynamic range by the LUT (Fig. 13) .
Since the digital filter used for generation of
the ultra-low range frequency component
becomes 100 or more taps, it uses an IIR (infinite-
duration impulse response) filter in considera-
tion of the hardware scale. To offset the phase
distortion of the IIR filter, computation is carried

QSeparation Between Flat Area and Edge Areal

Color Correlation = RMH •~ GMH + GMH •~ BMH + BMH •~ RMH

RMH ; medium/high frequency area of R

Fig. 10 Processing for Facial Expression Improvement GMH ; medium/high frequency area of G
BMH ; medium/high frequency area of B
Processing of face area and others is performed
simultaneously. High Correlation Edge Area
Low Correlation Flat Area
After each processing result gives weight with
face mask data, it is added. Fig. 11 Hyper-Sharpness Processing
 ―321―
Vol.64, No.5, 2001 Image Processing of Digital Minilab System

Fig. 12 Hyper-Sharpness Processing Block Diagram

luminance component (Y) of Medium/high frequency is divided into high frequency and Medium frequency, and a
gain is controlled by the color correlation value.

unlike the conventional lab system. For exam-

ple, you can get high quality images even use a
simple camera by using the Frontier's functions
for correcting lens distortion aberration, lateral
chromatic aberration and the ambient illumina-
tion level. As already discussed, image process-
Inclination on the light side (ƒ¿_light) and that on the dark
ing not only improves basic image quality, it also
side (ƒ¿_dark) are separately set.
enables performance to be maintained. Since
Reference Point YO: Reference Point of Dynamic
Range Compression
we are engaged in equipment development, we
Dynamic Range compression is applied toward will continue further studies in an effort to real-
this point.
ize optimum cost performance of the image sys-
Fig. 13 Hyper-Tone Processing
tem as a whole. The Minilab system is a print-
ing machine for production purposes, so any
image processing method that ensures the high-
est image quality cannot ignore printing capac-
ity and price. Therefore, it is important to
work in cooperation with those engaged in algor-
ithm development and implementation.
Although not mentioned in this report, there
have been a growing number of cases where
digital images are directly input due to the wide
spread use of digital cameras. So services in
this field are urgently needed. Thus, one of the
biggest problems facing the Minilab system is
Fig. 14 Hardware Configuration (Hyper-Tone Process- how to cope with this issue and provide satisfac-
ing) tory services.
To offset the phase distortion of the IIR filter, the
References
processing direction is reversed by SRAM.
1) Hiroaki NAKAMURA, "Digital Image Processing
for the Forntier 350 Digital Minilab" IS&T's 1999
out twice in the reverse directions.
PICS Conference
To downscale the hardware, this processing is 2) Tohru MATAMA, Fumihiro SONODA, and Hisao
also converted into one-dimensional processing, ARAI, "Digital Image Processing for a Photo-
and is performed twice in the horizontal and graphic Print System" ICPS '98
3) Yoshio OZAWA, Naoyuki MORITA, Kanenori
vertical directions after rotation in the image
OCHIAI, Tomonori NISHIO, Yoshinori MORI-
memory (Fig. 14). MOTO, Tadashi UEKUSA, Azuchi ENDO, Hiroaki
NAKAMURA, and Shinji ITO, "Development of
4. Conclusion Digital Minilab Systems Frontier 350 / 370" FUJI-
FILM RESERCH & DEVELOPMENT (No. 45-
A wide variety of image processing functions 2000)
can be installed in the digital Minilab system,