There has been much debate in the history of philosophy aimed at determining what it is, exactly, that makes a person who and what she is. Varying theories have offered a wide range of concepts in pursuit of the answer to this question. Some thinkers, such as B.F. Skinner, have claimed that it is observable behavior patterns that determine who and what a person is. Yet other thinkers, such as Carl Jung, have attributed unconscious motivators as being determinative in deciphering who and what a person is. Jung claims that it is the conscious and unconscious working together that determines who and what a person is. The purpose of this thesis is to discover evidence that supports or disproves the theory of self in which the unconscious and conscious work together to determine who and/or what a person is. This is done by semiotically analyzing the Visual Products (VP) of Visual Product Producers (VPP) who were or are afflicted with Bipolar Disorder. This thesis consists of the semiotic analysis of selected works by Jackson Pollock, Virginia Woolf, Vincent Van Gogh, and myself (Michael Main). Semiotic analysis studies how meanings are generated as opposed to what meanings are generated. It should be noted that semiotics was used strictly as a method of analysis and not as a guiding philosophy. In examining how the works of the selected VPPs generate meaning, it is hoped that evidence is produced that proves or disproves the theory of who or what a person is as determined by the interaction of the conscious and unconscious.

No description available.

No description available.

No description available.

No description available.

DNA Microarray is an innovative tool for gene studies in biomedical research. It is capable of testing and extracting the expression of large number of genes in parallel. Its applications can vary from cancer diagnosis to human identification. A DNA microarray experiment generates an image which has the genetic data embedded in it. Fast, accurate, and automatic routines for processing and compression of these images do not exist. For processing and modelling of micoarray images, we introduce a new, fast and accurate approach in this thesis. A new lossless compression method for microarray images is introduced that provides an average compression ratio of 1.89:1, and that outperforms other lossless compression schemes and the work of other researchers in this field. For the lossy compression, our new method has overcome the rate-distortion curve of JPEG. A new scanning method called spiral path, and a new spatial transform called C2S are introduced in this thesis for lossless and lossy compression of microarray images. / Thesis / Master of Applied Science (MASc)

Interactive deep learning frameworks are crucial for effectively exploring and analyzing complex image datasets in visual analytics. However, existing approaches often face challenges related to inference accuracy and adaptability. To address these issues, we propose ImageSI, a framework integrating deep learning models with semantic interaction techniques for interactive image data analysis. Unlike traditional methods, ImageSI directly incorporates user feedback into the image model, updating underlying embeddings through customized loss functions, thereby enhancing the performance of dimension reduction tasks. We introduce three variations of ImageSI, ImageSI$_{text{MDS}^{-1}}$, prioritizing explicit pairwise relationships from user interaction, and ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{PHTriplet}}$, emphasizing clustering by defining groups of images based on user input. Through usage scenarios and quantitative analyses centered on algorithms, we demonstrate the superior performance of ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{MDS}^{-1}}$ in terms of inference accuracy and interaction efficiency. Moreover, ImageSI$_{text{PHTriplet}}$ shows competitive results. The baseline model, WMDS$^{-1}$, generally exhibits lower performance metrics. / Master of Science / Interactive deep learning frameworks are crucial for effectively exploring and analyzing complex image datasets in visual analytics. However, existing approaches often face challenges related to inference accuracy and adaptability. To address these issues, we propose ImageSI, a framework integrating deep learning models with semantic interaction techniques for interactive image data analysis. Unlike traditional methods, ImageSI directly incorporates user feedback into the image model, updating underlying embeddings through customized loss functions, thereby enhancing the performance of dimension reduction tasks. We introduce three variations of ImageSI, ImageSI$_{text{MDS}^{-1}}$, prioritizing explicit pairwise relationships from user interaction, and ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{PHTriplet}}$, emphasizing clustering by defining groups of images based on user input. Through usage scenarios and quantitative analyses centered on algorithms, we demonstrate the superior performance of ImageSI$_{text{DRTriplet}}$ and ImageSI$_{text{MDS}^{-1}}$ in terms of inference accuracy and interaction efficiency. Moreover, ImageSI$_{text{PHTriplet}}$ shows competitive results. The baseline model, WMDS$^{-1}$, generally exhibits lower performance metrics.

Medical images are extremely complicated to comprehend for a person without expertise. The limited number of practitioners across the globe often face the issue of fatigue due to the high number of cases. This fatigue, physical and mental, can induce human-errors during the diagnosis. In such scenarios, having an additional opinion can be helpful in boosting the confidence of the decision-maker. Thus, it becomes crucial to have a reliable Visual Question Answering (VQA) system which can provide a "second opinion" on medical cases. However, most of the VQA systems that work today cater to real-world problems and are not specifically tailored for handling medical images. Moreover, the VQA system for medical images needs to consider a limited amount of training data available in this domain. In this thesis, we develop a deep learning-based model for VQA on medical images taking the associated challenges into account. Our MedFuseNet system aims at maximizing the learning with minimal complexity by breaking the problem statement into simpler tasks and weaving everything together to predict the answer. We tackle two types of answer prediction - categorization and generation. We conduct an extensive set of both quantitative and qualitative analyses to evaluate the performance of MedFuseNet. Our results conclude that MedFuseNet outperforms other state-of-the-art methods available in the literature for these tasks. / Master of Science / Medical images are extremely complicated to comprehend for a person without expertise. The limited number of practitioners across the globe often face the issue of fatigue due to the high number of cases. This fatigue, physical and mental, can induce human-errors during the diagnosis. In such scenarios, having an additional opinion can be helpful in boosting the confidence of the decision-maker. Thus, it becomes crucial to have a reliable Visual Question Answering (VQA) system which can provide a "second opinion" on medical cases. However, most of the VQA systems that work today cater to real-world problems and are not specifically tailored for handling medical images. In this thesis, we propose an end-to-end deep learning-based system, MedFuseNet, for predicting the answer for the input query associated with the image. We cater to close-ended as well as open-ended type question-answer pairs. We conduct an extensive analysis to evaluate the performance of MedFuseNet. Our results conclude that MedFuseNet outperforms other state-of-the-art methods available in the literature for these tasks.

Hardwood timber is a substantial economic staple in the eastern U.S., where primary hardwood processors produce more than 10 billion board feet of sawn hardwoods annually. There are over 3,500 sawmills producing hardwood lumber in the Southeastern portion of the United States. Present trends such as increasing labor costs and limited supplies of high quality logs have forced hardwood lumber manufacturers to increase their efforts to maximize the utilization of this raw material. In order to make money in such a competitive business, these sawmills must produce the highest possible grade of lumber from each saw log they process. Of all the primary and secondary processing procedures that are used to transform round wood into a final product, the sawmill edging and trimming operations have the most substantial effect on the grade and, hence, the value of the material produced. Currently, the grading of rough hardwood lumber is done manually by human inspectors according to standardized grading rules developed by the National Hardwood Lumber Association (NHLA, 1994). Standard hardwood edging and trimming operations are less than optimum because of the complexity of the grading rules, the complexity of the decision making processes involved, possible operator fatigue, and the imprecision with which lumber can be sawn by the available equipment. Studies have shown that there is a potential to increase hardwood lumber value by over 20 percent if optimum edging and trimming could be performed in hardwood sawmills. Even a small portion of this percentage would substantially increase the profit of hardwood lumber manufacturers. And this can be achieved just by utilizing some degree of automation. That is, some type of system must be designed that can scan a board to sense important hardwood features, make correct edging and trimming decisions, and then control down stream edgers and trimmers with minimal operator intervention. The most difficult part in the development of this automatic edging and trimming system is to get enough major defect information to make very good edging and trimming decisions. This thesis describes the research that was performed to build a prototype system that can collect images of boards and extract major defect information for making good edging and trimming decisions. The images that are collected include Black/White and laser profile images. Necessary defect information to be extracted for making edging and trimming decisions includes the location and size of large grading defects and areas of the board that are too thin. This thesis talks about the hardware that was used for collecting the needed board images. This includes a discussion of both the Black/White and laser profile imaging systems. The data collection boards that were used for transferring images from these imaging systems to computer memory are also described. This thesis also describes the computer vision algorithms that were developed to extract defect information needed for making improved edging and trimming decisions. Some of the processing steps involved include background extraction, both global and local segmentation, connected component labeling and small area elimination and merging. Processing results obtained of green red oak samples show that both hardware and software of the prototype system seem to work well. However, since the program needed to actually create the edging and trimming solution based on defect information found by the computer vision system was not available it was impossible to quantitatively determine the value improvement to proposed system might offer. / Master of Science

Many older documents are of poor quality and are deteriorating with the passage of time. Furthermore, many of these documents are currently stored in the form of paper or photographic film, and therefore storage space, accessibility, and security are serious problems. Because of these problems, many older documents are being restored and converted to digital form for archival. Image scanners which perform these tasks must be fast and must produce digital images that are of high quality. A Scan Optimizer was developed by Image Processing Technologies, Inc., to assist in the restoration and conversion processes. This device dramatically improves the throughput of image scanners by the use of high-speed complicated image processing. As originally designed, however, the IPT Scan Optimizer requires the manual adjustment of several parameters to obtain the best results for a given document. Because of this manual adjustment, human intervention was often required, and conversion speeds were drastically reduced. In order to remove the need for manual adjustments, an Automatic Parameter Setting (APS) algorithm was developed. This algorithm relies on histogram analysis over non overlapping image regions for the computation of scan parameters. This thesis describes a hardware realization of the APS algorithm. To achieve a real-time implementation that is low in cost, several compromises were necessary in the design. This two-board set is compatible with the AT bus, and is composed of one SPARC RISC chip, four Xilinx Programmable Logic Cell Arrays, ten PALs and many RAMs and supporting logic. This system has been designed, built, and successfully tested with many documents. / Master of Science