Research Interests
My early research work was concerned with shape recognition. I published several articles in the late 80’s and early 90’s mainly in the area of pattern recognition. The most noted publication is the comparative study about the moment invariants. This article is the most cited article among all of my publications. This article also won the prestigious most honorable mention award given by the international pattern recognition society for the year 1990.
My publications have been cited in several books, theses and more than 80 refereed international journals and conference proceedings.
During the last five years I have published 21 refereed journal and conference papers.
These papers cover many aspects of research dealing with image segmentation, image enhancement, image description, biomedical image processing and embedded system design.
Since joining GSU I distributed my research between digital image processing, pattern recognition and embedded system design.
I have been a co investigator in the state of Georgia Yamacraw team grant for development of Embedded Software and currently I am a co investigator in the p20 NIH bioinformatics grant. I have also filed four utility patent applications.
Currently I am supervising 3 PhD students and 2 M.Sc. working in my research area. The following sections highlight some of my research interests and goals:
Many researchers have used moments to compute wide range of functions particularly those ones that can be represented by series expansions. The moments are widely used in many areas. They have been used successfully to classify, recognize and segment objects in many pattern recognition and image processing applications.
The moments gained popularity in pattern recognition for their suitability to be normalized and become invariant against many parameters such as scale, shift, rotation, contrast, and blur. The regular moments can be considered the building block of all other related forms of moments such as orthogonal moments. This means that any gain or efficiency in computing the regular moments can be passed on to the other forms of moments. In fact the efficiency of computing the moments has been already used to improve the computation of other transforms such as DST and DCT.
We have introduced a new transformation procedure. This transformation is guaranteed to reduce the number of operations needed to compute a particular moment order of a one dimensional signal to at least one half and one fourth for computing a particular order of two dimensional signal or image.
The discrete cosine transform (DCT) is often used in signal
and image processing, especially for data compression. The excellent energy
compaction property of the DCT is the main reason for its popularity. We
introduced a new pre-processing approach for the removal of the blocking
effect, associated with the standard 2D-DCT,. In this
method, the processing of the image is started from the top corners by
zigzagging rows and columns of an image into horizontal, vertical or diagonal
streams of pixels. These pixel streams are transformed into one-dimensional
line arrays. The one-dimensional DCT transform is applied to each array. The
experimental results revealed the effectiveness of this method in reducing the
blocking effect associated with the standard DCT. The zigzag line method
produces very good subjective as well as PSNR results.
Image retrieval plays an important role in broad spectrum of applications. Content-based retrieval (CBR) is one of the popular choices in many biomedical and industrial applications. Discrete image transforms have been widely studied and suggested for many image retrieval applications. The Discrete Wavelet Transform (DWT) can be used to form the basis for extracting features for retrieving images based on the description of a particular object within the scene. In our research we highlight the common features between compression and retrieval. Several examples have been used to test the DWT retrieval system. A comparison between DWT and Discrete Cosine Transform (DCT) has been also made. The retrieval system using DWT requires preprocessing and normalization of images, which might slow down the retrieval process. The accuracy of the retrieval using DWT has been significantly improved by incorporating efficient K-Neighbor Nearest Distance (KNND) measure in our system.
Biofilms are biological microorganisms
attached to surfaces and develop a complex heterogeneous three-dimensional
structure. Understanding interactions in multispecies biofilms may
contribute to treatment of polymicrobial biofilm infections as well as rational design of engineered
biofilms.
Analysis of microbial biofilms by confocal laser scanning microscopy (CLSM) yields a stack of
digital images that can be combined to give a three-dimensional view of the biofilm. Information on the relationships among bacteria in
multispecies biofilms can
be obtained from CLSM image stacks by combining pattern recognition and image
processing techniques such as clustering and segmentation.
A multi resolution image analysis utilizing image segmentation of image contours as boundaries for extracting objects at multi resolution levels has been successfully implemented by our biofilm research group (Collaboration effort with Biology dept.). The scheme reduced the complexity of the segmentation problem considerably and simplified the clustering process.
4- Real
time image processing
One particular new area that is emerging in the last few years is digital image processing on GPU (Graphical Processing Unit). GPU image processing appears to be a promising platform to increase the capabilities of our image processing algorithms and achieve real time image processing. Development and applying new algorithms to take advantage of the immense processing power of GPUs is a major focus of this research, particularly applications that utilize GPUs in other fields such as biology and medicine.