INFORMATION RETRIEVAL FROM IMAGES DEGRADED BY FILM GRAIN NOISE
THESIS SUBMITTED TO
THE INDIAN INSTITUTE OF TECHNOLOGY DELHI FOR AWARD OF THE DEGREE OF
DOCTOR OF PHILOSOPHY
BY
ANIL SHRINIVAS TAVILDAR
DEPARTMENT OF ELECTRICAL ENGINEERING
INDIAN INSTITUTE OF TECHNOLOGY, DELHI
JULY 1984
6p,
'ON
Irt-(ARJ4?
CERTIl'ICATE
This is to certify that the dissertation, "Information Retrieval from images Degraded by Film Grain Noise," which is being submitted by Anil Shrinivas Tavildar for the award of the degree of Doctor of Philosophy to the Indian Institute of Technoloa, New Delhi, is a record of bonafide research work.
This dissertation has reached the standard fulfilling the requirements of regulations relating to the degree. The results obtained in this dissertat- a have not been submitted to any
other University or Institute for the award of any degree or diploma.
11.1i. GUPTA Assistant Professor
Department of Electrical Engineering
Indian Institute of Technelogy,Delhi
New Delhi 110 016, India
S.N. GUPTA Professor
ACKEOVILMGRIENT
I wish to express my deep sense of gratitude to Professor S.N. Gupta, and Dr. H.M. Gupta for their valuable guidance and constant encouragement during the course of this research work.
I am also grateful, to the Secretary, Department of Electronics,
Government of India, for
encouraging and
permitting me to pursue the research work.My sincere thanks are due to Dr.(Mrs.) S. ChitaMbar of AIMS, Mr. Venugopal and Mr. P. Dhar of Space Application Centre, ISRO, Ahmedabad, for providing access to technical information and the library facilities. A word of thanks is due to my friends
Dr. K.S.K. Sai, Dr. S.P. Uttam, Dr. P. Mohana Shankar and
Dr. V.K. Jain for their support - both in academic and non-
academic matters - which has been instrumental in reducing the rigors of doctoral research.
I wish to express my sincere
appreciation and thanks to Mr. P.M. Padmanabhan Nambiar
for his exce)lent typing. I grate- fully acknowledge thetimely assistance provided by
Mr.SuchendraKulkarni during
the preparation of the thesis.Finfilly, a word of praise
to my wife Archana,
for herforbearance and co-operation without which this
work could nothave been possible.
ABSTRACT
Digital image processing has acquired a high degree of impor- tance due to its applications in many diverse fields. The image is commonly recorded on a photographic film. The information retrieval is effected by various factors like the random noise, the system
nctlinearities, the blurring effects etc. during the recording. ond the scanning of images.
Any developed photographic film shows a random grainy
appearance which constitutes film grain noise. The thesis presents results of studies on information retrieval from photographic
images in the presence of film grain noise. The information retrieval techniques are based on decision theoretic concepts.
An existing model of film grain noise has been modified to include the effect of grain clustering. Binary detection strate- gies hove been developed for films with and without grain cluster- ing. A developed film is scanned using a top-hat square aperture.
It is assumed be positioned. exactly over one pixT1 at a time.
Blurring degradations have not been considered.
The detection strategy is based on multiple observations.
For films without grain clustering, the analysis is applicable for the entire range of signal dependency parameter. It has been shown that the use of decision theoretic concepts leads to a supe- rior strategy as compared to that with the ad-hoc threshold.
The effect of the light scattering in the emulsion has been considered as a source of intersymbol interference and the binary
detection has also been studied in the presence of film grain noise and ISI. The performance has been evaluated for two films.
T-ne detection strategy has been used for retrieval of infor- Dation from the low contrest soft-tissue binary X-ray images in the presence of film grain noise. It has been shown that the strategy based on multiple samples results in a significant increase in tile probability of detection or true positive identification of the malignant tissues.
The MIT estimation algorithms have been proposed for images
,,,rith continuously varying grey levels. Considering the large COM- putational effort involved in the implementation of the MAP algo-
rithms, a signal independent transformation has been developed Irhen the signal dependency parameter, equals 1/3. The MAP algo- rithm has been modified using the proposed SIT. The performance has been evaluated in terms of the normalized mean square error and the efficiency of various estimation algorithms has been discussed.
Linear mean square error (LOSE) filters (Wiener filters) have been developed considering the film grain noise to be signal inde- pendent multiplicative noise in the intensity domain. The mean square error has been evaluated for binary and multilevel images by assuming the signal spectra corresponding to Laplacian autocorrela- tion function and gaussian spectra respectively. The effect of the varying noise-ice-signEl bandwidth ratio has been examined«
CONTENTS
Page ABSTRACT
LIST OF EBREVIATIONS LIST OF SYMBOLS
LIST OF FIGURES
iii
iv
viii
LIST OF TABLES
CHAPTER 1 INTRODUCTION 1
1.1 Photograp:pto Image Recording 3
1.2 Image Imp aLcraents 3
1.3 Film Grain ibise 5
CHAPTER 2 PHOTOGRAPHIC PROCESS, GRA'ilULARITY
MID 7
INFORMATION RISTRE VAL
2.1 Models for Image Formation 7
2.2 rhotoc,fraphic Process 9
2.3 Inpu.t-Ou.-bput Relationship 11
2.4 Scannirg Effects 17
2.5 Film Grain Noise and Granularity 24
2.5 . / Granularity Model s 27
2.6 Information Retrieval. Techniques 36
CHAPTER 3 MODJS FOR PiLM GRAIN NOISE 49
3.1 Huang Model 50
3.2 Yu Mod. el 54
Poge 3.3 hederi-Samchuk Model for Imaging System 55
3.4 Grain Clustering Nodel 58
CHAPTER 4 BIK1RY DETECTION IN FILM GRAIN NOISE 68 4.1 Detection in Film Grain Noise 69
4.1.1 Results and Discussion 75
4.2 Detection in Film Grain ;Poise with 85 Grain Clustering
4.2.1 Performance of Detection in Presence of 88 Clustering
CHAPTER 5 BMRY DETErTION OF RADIOGRAPHIC 92 IN FILM GRAiN NOISE
5.1 X-ray Imaging Processes 92
5.2 Binary Detection 98
5.3 Discussion 102
CHAPTER 6 BINARY DETECTION IN FILM CRAIN NOISE AND 110 INYERSYMBOL INTERFERENCE
6.1 Intersymbol Interference 110
6.2 Detection in Film Grain Noise and 114 Intersymbol Interference
6.3 Discussion 127
CHAPTER 7 MAP ESTIUTION IN EiLM GRAIN NOISE 129 7.1 Estimation in Signal Dependent Film 130
Grain Noise
