Image processing--Digital techniques

Digital videos and images are effective media for capturing spatial and ternporal
information in the real world. The rapid growth of digital videos has motivated
research aimed at developing effective algorithms, with the objective of obtaining useful
information for a variety of application areas, such as security, commerce, medicine,
geography, etc. This dissertation presents innovative and practical techniques, based on
statistics and machine learning, that address some key research problems in video and
image analysis, including video stabilization, object classification, image segmentation,
and video indexing.
A novel unsupervised multi-scale color image segmentation algorithm is proposed.
The basic idea is to apply mean shift clustering to obtain an over-segmentation, and
then merge regions at multiple scales to minimize the MDL criterion. The performance
on the Berkeley segmentation benchmark compares favorably with some existing approaches.
This algorithm can also operate on one-dimensional feature vectors representing
each frame in ocean survey videos, which results in a novel framework for building
a hierarchical video index. The advantage is to provide the user with the flexibility
of browsing the videos at arbitrary levels of detail, which makes it more efficient for users to browse a long video in order to find interesting information based on the
hierarchical index. Also, an empirical study on classification of ships in surveillance
videos is presented. A comparative performance study on three classification algorithms is
conducted. Based on this study, an effective feature extraction and classification algorithm
for classifying ships in coastline surveillance videos is proposed. Finally, an empirical
study on video stabilization is presented, which includes a comparative performance study
on four motion estimation methods and three motion correction methods. Based on this
study, an effective real-time video stabilization algorithm for coastline surveillance is
proposed, which involves a novel approach to reduce error accumulation.

The development of an unmanned underwater vehicle at Florida Atlantic
University with onboard optical sensors has prompted the temporal and spatial optical
characterization of Port Everglades, with in-situ measurements of the turbidity,
conductivity, and temperature. Water samples were collected for laboratory analysis
where attenuation and absorption were measured with a bench top spectrometer. All of
the measurements showed a high degree of variability within the port on a temporal and
spatial basis. Correlations were researched between the measured properties as well as
tide and current. Temporal variations showed a high correlation to tidal height but no
relation was found between turbidity and current, or salinity. Spatial variations were
primarily determined by proximity to the port inlet. Proportionality constants were
discovered to relate turbidity to scattering and absorption coefficients. These constants
along with future turbidity measurements will allow the optimization of any underwater
camera system working within these waters.

Lower prices of video sensors, security concerns and the need for better and faster
algorithms to extract high level information from video sequences are all factors which
have stimulated research in the area of automated video surveillance systems. In the
context of security the analysis of human interrelations and their environment provides
hints to proactively identify anomalous behavior. However, human detection is a
necessary component in systems where the automatic extraction of higher level
information, such as recognizing individuals' activities, is required. The human detection
problem is one of classification. In general, motion, appearance and shape are the
classification approaches a system can employ to perform human detection. Techniques
representative of these approaches, such us periodic motion detection, skin color
detection and MPEG-7 shape descriptors are implemented in this work. An infrastructure
that allows data collection for such techniques was also implemented.

The focus of this research is on images extracted from surveillance videos that
have a low resolution and are taken under low illumination. In recent years, great
advances have been made in face recognition and many studies mention results of 80%
and 90% of recognition efficiency, however, most of these studies reported results using
face images under controlled conditions. Current surveillance systems are equipped with
low resolution cameras and are located in places with changing illumination, as opposed
to a controlled environment. To be used in face recognition, images extracted from
videos need to be normalized, enlarged and preprocessed. There is a multitude of
processing algorithms for image enhancement, and each algorithm faces its advantages
and disadvantages. This thesis presents a novel method for image enlargement of human
faces applied to low quality video recordings. Results and comparison to traditional
methods are also presented.

Video compression technology promises to be the key to the transmission of motion video. A number of techniques have been introduced in the past few years, particularly that developed by the Motion Picture Experts Group (MPEG). The MPEG algorithm uses Motion Estimation to reduce the amount of data that is stored for each frame. Motion Estimation uses a reference frame as a codebook for a modified Vector Quantization process. While an exhaustive search for Motion Estimation Vectors is time-consuming, various fast search algorithms have been developed. These techniques are surveyed, and the theoretical framework for a new search algorithm is developed: Densely-Centered Uniform P-Search. The time complexity of Densely-Centered Uniform P-Search is comparable to other popular Motion Estimation techniques, and shows superior results on a variety of motion video sources.

A user interface that has objects familiar to the user will be easier to use. In this thesis, a user interface that is customizable to any color bitmap is proposed. The most significant problem with this approach is the problem of finding objects in a color bitmap. A solution to the problem is proposed and evaluated using an analysis tool, developed for this thesis, called Workbench. Current image detection methods are evaluated and compared to the solution proposed using Workbench. The proposed solution is then evaluated for the YIQ and HSI color mappings. The results of this investigation and recommendations for future work is proposed.

This thesis presents an image coding system using binomial QMF based subband decomposition and vector quantisation. An attempt was made to compress a still image of size 256 x 256 represented at a resolution of 8 bits/pixel to a bit rate of 0.5 bits/pixel using 16 channel subband decomposition with binomial QMFs and coding the subbands using a full search LBG Vector Quantizer (VQ). Simulations were done on SUN work station and the quality of the image was evaluated by computing the Signal to Noise Ratio (SNR) between the original image and the reconstructed image.

Optical Character Recognition systems have many applications in today's world of electronic computing. Various software implementations are currently being used. This thesis evolves a massively parallel hardware implementation for the system that is VLSI scaleable and may lead to substantial increase in the processing speed. This system involves various stages for preprocessing and processing of the image implemented with SIMD architecture, using simple processing elements and near neighbor communications. The architecture evolved is simulated using the Verilog Hardware Description Language. This project should provide a framework for a massively parallel processing architecture for such systems. It is expected that this project will lead to the design and implementation of a real time system.

In this thesis, a partial fraction expansion of a separable-in-denominator 2-D transfer function is given. Based on this expansion, several novel realizations of separable-in-denominator 2-D filter are provide. These realizations have the properties of highly parallel structure and improved throughput delay. The performance figures are given in the tables. A method of evaluation of quantization error of separable-in-denominator 2-D filter is also derived by using the residue method. Formulas for calculation of roundoff noise of proposed structures are provided. Two programs which can be used to calculate the roundoff noise of proposed structure are listed in the Appendix. To run the programs, we need only to input the constant coefficients of expanded transfer function. At last, an optimal block realization of separable-in-denominator 2-D filter is discussed and the criterion for the absence of limit cycles for a second-order 2-D block is given.

A barrier to the use of digital imaging is the vast storage requirements involved. One solution is compression. Since imagery is ultimately subject to human visual perception, it is worthwhile to design and implement an algorithm which performs compression as a function of perception. The underlying premise of the thesis is that if the algorithm closely matches visual perception thresholds, then its coded images contain only the components necessary to recreate the perception of the visual stimulus. Psychophysical test results are used to map the thresholds of visual perception, and develop an algorithm that codes only the image content exceeding those thresholds. The image coding algorithm is simulated in software to demonstrate compression of a single frame image. The simulation results are provided. The algorithm is also adapted to real-time video compression for implementation in hardware.