2 STILL IMAGE S COMPRESSIO N
as compressio n i n signa l o r imag e processing . Le t u s mentio n analysis , diagnosti c
feature extractio n an d scen e interpretation . W e wil l agai n mee t thes e problem s i n
Chapter 3 wher e frequenc y modulate d signal s nee d t o b e extracte d fro m a nois y
environment.
A few othe r word s on th e digita l revolutio n wil l help i n introducing th e debate .
Then th e scientifi c issue s raise d b y these ne w technologies wil l be illustrate d b y th e
specific exampl e o f deblurring th e picture s o f the supernov a SN19S7A take n b y th e
Hubble telescope .
The digita l revolutio n ha s a growin g impac t o n ou r everyda y life . Indee d thi s
revolution i s modifyin g ou r perceptio n o f th e surroundin g worl d an d i s deepl y af -
fecting ou r relatio n t o others . Th e digita l revolutio n i s speedin g communicatio n
while i t i s introducin g som e systemati c bia s i n th e delivere d messages .
This revolutio n spran g u p i n the earl y fortie s muc h before adequat e technolog y
was available . Th e scientist s wh o pioneere d thi s endeavou r wer e eithe r mathemati -
cians o r physicists . Amon g th e mathematicians , le t u s lis t Joh n vo n Neumann ,
Claude Shanno n an d Norber t Wiener . Amon g th e physicist s w e mee t Denni s Ga -
bor an d Eugen e Wigner . The y share d a vision, a drea m wher e mathematica l logic s
started a fruitfu l interactio n wit h th e neurophysiolog y o f th e brain , wher e statis -
tical mechanic s coul d b e applie d t o improvin g telecommunicatio n an d t o bette r
understand th e hidde n structur e o f th e huma n language .
This interdisciplinar y progra m too k plac e i n th e 'Institut e fo r th e Unit y o f
Science'. Thi s institut e neve r existe d a s a building . I t wa s a n institut e withou t
walls. I t wa s a n endeavour . Thi s endeavou r i s transformin g ou r everyda y life .
Today th e image s o f th e world , th e landscapes , th e picture s o f th e citie s a s wel l
as th e voice s o f thei r inhabitants , th e sounds , th e music , amon g othe r image s an d
signals, ar e transmitte d a s binar y sequences .
One migh t immediatel y objec t tha t thi s canno t mak e an y sense . Indee d th e
surrounding worl d i s to o complex , ric h an d dee p t o b e reduce d t o th e limite d siz e
of finite sequence s o f 0 or 1. On e canno t avoi d som e downsizin g o r shrinkin g a s i n
these fairy-tale s wher e a gian t i s entering a smal l bottle.. . Thi s shrinkin g i s name d
'compression' an d i s a t th e hear t o f ou r discussion .
In man y example s a give n signa l o r imag e (whic h wil l be denote d b y / ) ca n b e
written a s a su m u + v betwee n a 'hig h priority ' componen t u an d a n 'lo w priority '
item v wher e v i s ofte n corrupte d b y som e additiv e noise . W e expec t thi s hig h
priority componen t u t o b e wel l structured. Thi s i n principle implie s tha t u ca n b e
compressed. I n contras t th e v componen t i s les s structure d an d ver y ofte n ther e
are n o adequat e model s fo r describin g v. A n exampl e o f suc h a v componen t i s
the hai r o f a perso n whos e fac e i s bein g portraited . The n a n idea l compressio n
algorithm i s the on e tha t woul d focu s o n modelin g u. I n th e exampl e o f a portrait ,
the painte r wil l us e a kin d o f modelin g o f th e hai r instea d o f paintin g eac h singl e
hair separetely . I n imag e processin g on e i s lookin g fo r statistica l model s o f th e
textured component . W e wil l retur n t o suc h u + v model s i n Sectio n 11and 13.
Giving some high priority t o some information whil e deleting some other dat a is
a difficul t an d ofte n painfu l decision . I t shoul d b e based o n a full a priori knowledg e
about wha t i s goin g on . Suc h a decisio n immediatel y lead s t o a kin d o f viciou s
circle wheneve r th e sourc e o f ou r knowledg e i s precisel y wha t i s transmitte d o r
compressed. Thes e remark s rais e th e issu e o f finding a criterio n fo r decidin g wha t
defines th e u component insid e a given signal / . Thi s requires a scientific knowledg e
which i n mos t o f th e case s i s no t available .
Previous Page Next Page