«Characterization of Sun Glitter Statistics in Ocean Video A NISE funded Basic Research Project Katie Rainey Eric Hallenborg Approved for public ...»
Technical Report 2031
Characterization of Sun Glitter
Statistics in Ocean Video
A NISE funded
Basic Research Project
Approved for public release.
San Diego, CA 92152-5001
San Diego, California 92152-5001
J. J. Beel, CAPT, USN C. A. Keeney
ADMINISTRATIVE INFORMATIONThis report was prepared by the ISR/IO Department (Code 56), SPAWAR Systems Center Pacific (SS Pacific), San Diego, CA. The work is funded by the Naval Innovative Science and Engineering (NISE) Program at SSC Pacific as a Basic Research project.
Released by Under authority of H. Buck, Head D. Holifield, Head Advanced Analysis ISR Division Systems Branch This is a work of the United States Government and therefore is not copyrighted. This work may be copied and disseminated without restriction.
Canon® is a registered trademark of Canon, Inc.
MATLAB® is a registered trademark of The MathWorks, Inc.
RESULTSA MATLAB® tool has been developed with which to analyze several key statistics of the collected video. This report details the functionality of this tool and demonstrates the behavior of a number of spatial-, temporal-, and spectral-domain statistics of glitter video. A detailed background on previous glitter-related research is also given.
RECOMMENDATIONSAdditional research is required to fully determine the possible applications of this work. Data collec- tions in a more controlled environment would enable better understanding of the impact of various collec- tion variables on statistical behavior.
iii CONTENTS EXECUTIVE SUMMARY
2. PREVIOUS WORK
2.1 EARLY STUDIES OF GLITTER
2.2 LEGACY OF COX AND MUNK
2.3 PHOTOGRAPHIC DATA COLLECTIONS
2.4 INTRODUCTORY REFERENCES TO GLITTER
3. DATA COLLECTION AND PROCESSING
3.1 VIDEO CAPTURE
3.2 GRAPHICAL USER INTERFACE
3.2.1 Data Processing
3.2.2 Data Visualization
3.2.3 Additional Functionality
4. DATA ANALYSIS
4.1 SET-UP AND DATA ACQUISITON
4.2 GLITTER STATISTICS
4.2.1 Time Domain Statistics
4.2.2 Power Spectral Density
4.2.3 Glitter Counts
4.2.4 Best-Fit Analysis
1. Screen shot of the MATLAB GUI for glitter analysis
2. Frame from DV14
3. Boundaries of nine selected ROIs
4. Time series from three regions
5. Time graphs in 16-by-128-pixel ROI
6. Time graphs in 16-by-16-pixel ROI
7. Time graphs in 1-by-16-pixel ROI
8. Spatial means in Sparkles region
9. PSD estimates at single pixels
10. PSD estimates in Flat region
11. PSD estimates in Loops region
12. PSD estimates in Sparkles region
13. Video frame clipped at various threshold levels
14. Glitter count time series and histogram
15. PSDs and best-ﬁt curves from three regions
16. Effects of pre-processing on PSDs and best-ﬁt curves
17. PSDs and best-ﬁt curves in three separate color channels
1. INTRODUCTIONGlitter analysis through image processing goes back at least to 1951 when Charles Cox and Walter Munk took photographs of sun glitter patterns over the Paciﬁc Ocean using cameras in the bomb bay of a World War II surplus B-17G aircraft. They derived a wave-slope probability density function for the glitter patterns by comparing the photographic density to the probability of a sun glitter wave slope. In the subsequent decades, further analysis of the geometric and statistical properties of glitter has been performed using still images of sun glitter as well as glitter synthesized with light sources such as lasers. Such analysis can reveal information about ocean roughness and wind speed. Glitter is often considered a hindrance to analysis of ocean scenes and glitter removal algorithms are often employed in pre-processing of ocean imagery. However, as demonstrated over years of research, glitter statistics can also provide insight into the properties and motion of the sea surface.
This research project is an empirical study of the spatial as well as temporal characteristics of sun glitter reﬂected on the ocean surface, and the spatial variability of statistical measures (such as temporal power spectral density) driven by spatial variations in environmental conditions. Glitter is observed through shore-based video of the ocean collected at various wavelengths (visible and infrared). The statistical properties of glitter can vary at each pixel in a video according to many factors, including surface wave conditions, wind speed, biological surfactants, and inﬂuence of nearby vessels. This research has numerous potential applications for intelligence, surveillance, and reconnaissance (ISR), such as for port monitoring and other surveillance systems. Research into glitter statistics will be succeeded by algorithm development in support of these various applications.
The research phase of this project consists of statistical analysis of an initial data collection of ocean video. Data collected under a variety of conditions will provide greater understanding of the effect of the surveillance system on glitter observations. Additional data will be collected to incorporate ocean behavior models into the statistical analysis. As the behavior of the collected data is better understood, so is our ability to develop automated detection tools for improved ISR and overall maritime domain awareness.
1 2. PREVIOUS WORK2.1 EARLY STUDIES OF GLITTER
The behavior of light sparkling on water, often referred to as glitter, has long been a subject of interest to artists and scientists alike. Many authors have examined glitter and considered relationships between glitter geometry and the surface slopes of the body of water. In 1822, a letter was written by Spooner  reporting measurements of the width of the glitter pattern in the Tyrrhenian Sea, indicating a maximum surface slope of 25 degrees. Measurements of glitter patterns over the Black Sea were recorded by Shuleikin . Many early qualitative descriptions of light reﬂected on water exist as well [3, 4].
Photographs have for many years been used to remotely sense ocean behavior. Experiments performed by researchers at the Naval Research Laboratory (NRL) and documented by Hulburt  studied the effects of polarized light on the ability to view the horizon or objects on the ocean surface. They used polarizing attachments to a sextant and binoculars to improve the performance of those instruments, and recorded the width of glitter pattern (as determined by photographs) versus sun elevation and wind velocity. The seminal photographic glitter study was conducted by Cox and Munk , with funding from the Air Force Research Center [see also 7–9]. The authors collected aerial photographs over the open ocean and deduced that the distribution of wave slopes is approximately Gaussian.
Motivated by the work of Cox and Munk  and Hulburt , NRL scientist Schooley  collected photographs of glitter patterns resulting from a ﬂash bulb rather than the sun (“ﬂash-sparkle pictures”).
Quantitative information was extracted from the pictures by computing the area of glitter which appears in square regions of the photographs and using the areas to construct a wave-slope probability, which is shown (as in ) to be approximately Gaussian. The standard deviation of the wind slope is plotted against wind velocity. Data from  was shown to agree with the ﬂash-sparkle data. Work by Barber  used a photographic set-up to measure wave direction by the correlogram. Later work from NRL [12, 13] developed a technique for determining the directional energy spectrum by means of an optical Fourier transform.
2.2 LEGACY OF COX AND MUNK
Largely inspired by the work of Cox and Munk, many theoretical studies have been written on the behavior of glitter and how it can be exploited in imagery. A series of papers by Longuet-Higgins [14, 15, 16] provides a geometric analysis of the creation and annihilation of points of light (“twinkles”) on a moving surface. This work continues to be relevant; studies of the statistics of specular point phenomenology, parallelling the work of Longuet-Higgins, were published decades later by Akhmedov, Gardashov, and Shifrin  and Gardashov .
A number of authors have continued the analysis of Stilwell  to determine the wave spectrum from a photograph. First- and higher-order analyses of spectrum computation from an optical Fourier transform under various sky conditions were given by Kasevich, Tang, and Henriksen  and Kasevich . An alternate approach to that taken by Stilwell  is proposed by Peppers and Ostrem . This new approach is not limited by small wave slope approximation, and is developed for several sky radiance models. A report by Bjerkaas and Riedel  out of The Johns Hopkins University Applied Physics Laboratory (JHU/APL) develops a new model for the elevation spectrum of wind-generated ocean waves, modifying, simplifying, and correcting errors from the models developed at NASA by Pierson and Stacy  and Pierson . The work of Stilwell  is extended by several authors from JHU/APL in [25, 26] and Chapman and Irani .
2 A procedure for estimating unpolarized irradiance reﬂectance and glitter patterns as a function of lighting and wind speed was given by Preisendorfer and Mobley [28, 29]. Tse, McGill, and Kelly  generated simulated whitecap and glitter radiance images, motivated by a need to remove effects of whitecaps or glitter, or to use as ground truth to verify procedures for deducing wave slopes, using the wave spectrum developed by Bjerkaas and Riedel . Zeisse [31, 32]—Reports out of the Naval Command, Control and Ocean Surveillance Center (NCCOSC—now SSC Paciﬁc).  derives an integral equation predicting slope distribution on the horizon, extending the work of .  documents FORTRAN code based on the Cox–Munk model to predict the radiance of the ocean surface. A later SSC San Diego (now SSC Paciﬁc) report also by Zeisse  studies grazing optical reﬂectivity over capillary waves. In work produced through the Naval Command, Control and Ocean Surveillance Center (NCCOSC) and SSC San Diego (both now SSC Paciﬁc), Zeisse [31, 32, 33] derived an integral equation predicting slope distribution on the horizon, extending the work of Cox and Munk ; documented FORTRAN code based on the Cox–Munk model to predict the radiance of the ocean surface; and studied grazing optical reﬂectivity over capillary waves. Elfouhaily, Chapron, Katsaros, and Vandemark  developed a new analytical spectrum model featuring wave age dependency in both long- and short-wave formulations, using data from the Joint North Sea Wave Project (JONSWAP)  to formulate the model. That work also summarizes and addresses shortcomings of some earlier spectrum models, including that proposed in .
Two English-language research groups have continued publishing glitter studies into the 21st century.
´ Alvarez-Borrego and collaborators have published a number of papers in the past three decades [36–45].
Relationships have been derived between the autocorrelations of surface wave hights and glitter patterns;
these relationships are then inverted so that wave heights can theoretically be obtained from aerial photographs of glitter patterns. Wave height spectra are then calculated via Fourier transform. The theory is compared to experimental data. The various papers cited present derivations in one and two dimensions and under other various conditions, and conditions under which inversion is possible are discussed. A series of papers by Cureton [46–49] corrects the model from , applies the model to real data, and extends the theory to a higher order. Additionally, Weber  gives theoretical discussion of an imaging system for viewing objects underwater through rough seas (most citations are in Russian).
2.3 PHOTOGRAPHIC DATA COLLECTIONS
Many photographic experiments and measurement collections have been made since the work of Cox and Munk and have been used to further glitter analysis. Hughes and Grant  took a collection of photographs, wave slopes, and other measurements taken of a ship wake in “dead water.” Theoretical analyses of this scenario are given by Hughes . The spectrum derivation with simpliﬁed calculations is detailed by Gotwols and Irani . Lubard, Krimmel, and Thebaud  determined wave number spectra and space-time spectra from optical video data collected during the West Coast Experiment [see 55] conducted by the Jet Propulsion Laboratory (JPL) at the Naval Ocean Systems Center (NOSC—now SSC Paciﬁc).