Learning Embeddings for Indexing, Retrieval, and Classification, with Applications to Object and Shape Recognition in Image Databases

Nearest neighbor retrieval is the task of identifying, given a database of objects and a query object, the objects in the database that are the most similar to the query. Retrieving nearest neighbors is a necessary component of many practical applications, in fields as diverse as computer vision, pattern recognition, multimedia databases, bioinformatics, and computer networks. At the same time, finding nearest neighbors accurately and efficiently can be challenging, especially when the database contains a large number of objects, and when the underlying distance measure is computationally expensive. This thesis proposes new methods for improving the efficiency and accuracy of nearest neighbor retrieval and classification in spaces with computationally expensive distance measures. The proposed methods are domain-independent, and can be applied in arbitrary spaces, including non-Euclidean and non-metric spaces. In this thesis particular emphasis is given to computer vision applications related to object and shape recognition, where expensive non-Euclidean distance measures are often needed to achieve high accuracy. The first contribution of this thesis is the BoostMap algorithm for embedding arbitrary spaces into a vector space with a computationally efficient distance measure. Using this approach, an approximate set of nearest neighbors can be retrieved efficiently - often orders of magnitude faster than retrieval using the exact distance measure in the original space. The BoostMap algorithm has two key distinguishing features with respect to existing embedding methods. First, embedding construction explicitly maximizes the amount of nearest neighbor information preserved by the embedding. Second, embedding construction is treated as a machine learning problem, in contrast to existing methods that are based on geometric considerations. The second contribution is a method for constructing query-sensitive distance measures for the purposes of nearest neighbor retrieval and classification. In high-dimensional spaces, query-sensitive distance measures allow for automatic selection of the dimensions that are the most informative for each specific query object. It is shown theoretically and experimentally that query-sensitivity increases the modeling power of embeddings, allowing embeddings to capture a larger amount of the nearest neighbor structure of the original space. The third contribution is a method for speeding up nearest neighbor classification by combining multiple embedding-based nearest neighbor classifiers in a cascade. In a cascade, computationally efficient classifiers are used to quickly classify easy cases, and classifiers that are more computationally expensive and also more accurate are only applied to objects that are harder to classify. An interesting property of the proposed cascade method is that, under certain conditions, classification time actually decreases as the size of the database increases, a behavior that is in stark contrast to the behavior of typical nearest neighbor classification systems. The proposed methods are evaluated experimentally in several different applications: hand shape recognition, off-line character recognition, online character recognition, and efficient retrieval of time series. In all datasets, the proposed methods lead to significant improvements in accuracy and efficiency compared to existing state-of-the-art methods. In some datasets, the general-purpose methods introduced in this thesis even outperform domain-specific methods that have been custom-designed for such datasets.

Isoperimetric Graph Partitioning for Data Clustering and Image Segmentation

Spectral methods of graph partitioning have been shown to provide a powerful approach to the image segmentation problem. In this paper, we adopt a different approach, based on estimating the isoperimetric constant of an image graph. Our algorithm produces the high quality segmentations and data clustering of spectral methods, but with improved speed and stability.

The Graph Analysis Toolbox: Image Processing on Arbitrary Graphs

Office of Naval Research (N00014-01-1-0624)

Self-Organizing Hierarchical Knowledge Discovery by an ARTMAP Image Fusion System

Classifying novel terrain or objects front sparse, complex data may require the resolution of conflicting information from sensors working at different times, locations, and scales, and from sources with different goals and situations. Information fusion methods can help resolve inconsistencies, as when evidence variously suggests that an object's class is car, truck, or airplane. The methods described here consider a complementary problem, supposing that information from sensors and experts is reliable though inconsistent, as when evidence suggests that an object's class is car, vehicle, and man-made. Underlying relationships among objects are assumed to be unknown to the automated system or the human user. The ARTMAP information fusion system used distributed code representations that exploit the neural network's capacity for one-to-many learning in order to produce self-organizing expert systems that discover hierarchical knowledge structures. The system infers multi-level relationships among groups of output classes, without any supervised labeling of these relationships.

CEDI: A Neural Model of Colour Vision, with Applications to Image Processing and Classification

Air Force Office of Scientific Research (F49620-01-1-0423); National Geospatial-Intelligence Agency (NMA 201-01-1-2016); National Science Foundation (SBE-035437, DEG-0221680); Office of Naval Research (N00014-01-1-0624)

Fixational Instability and Natural Image Statistics: Implications for Early Visual Representations

Under natural viewing conditions small movements of the eye, head, and body prevent the maintenance of a steady direction of gaze. It is known that stimuli tend to fade when they a restabilized on the retina for several seconds. However; it is unclear whether the physiological motion of the retinal image serves a visual purpose during the brief periods of natural visual fixation. This study examines the impact of fixational instability on the statistics of the visua1 input to the retina and on the structure of neural activity in the early visual system. We show that fixational instability introduces a component in the retinal input signals that in the presence of natural images, lacks spatial correlations. This component strongly influences neural activity in a model of the LGN. It decorrelates cell responses even if the contrast sensitivity functions of simulated cells arc not perfectly tuned to counterbalance the power-law spectrum of natural images. A decorrelation of neural activity at the early stages of the visual system has been proposed to be beneficial for discarding statistical redundancies in the input signals. The results of this study suggest that fixational instability might contribute to establishing efficient representations of natural stimuli.

A Fast BCS/FCS Algorithm for Image Segmentation

A fast and efficient segmentation algorithm based on the Boundary Contour System/Feature Contour System (BCS/FCS) of Grossberg and Mingolla [3] is presented. This implementation is based on the FFT algorithm and the parallelism of the system.

The What-And-Where Filter: A Spatial Mapping Neural Network for Object Recognition and Image Understanding

The What-and-Where filter forms part of a neural network architecture for spatial mapping, object recognition, and image understanding. The Where fllter responds to an image figure that has been separated from its background. It generates a spatial map whose cell activations simultaneously represent the position, orientation, ancl size of all tbe figures in a scene (where they are). This spatial map may he used to direct spatially localized attention to these image features. A multiscale array of oriented detectors, followed by competitve and interpolative interactions between position, orientation, and size scales, is used to define the Where filter. This analysis discloses several issues that need to be dealt with by a spatial mapping system that is based upon oriented filters, such as the role of cliff filters with and without normalization, the double peak problem of maximum orientation across size scale, and the different self-similar interpolation properties across orientation than across size scale. Several computationally efficient Where filters are proposed. The Where filter rnay be used for parallel transformation of multiple image figures into invariant representations that are insensitive to the figures' original position, orientation, and size. These invariant figural representations form part of a system devoted to attentive object learning and recognition (what it is). Unlike some alternative models where serial search for a target occurs, a What and Where representation can he used to rapidly search in parallel for a desired target in a scene. Such a representation can also be used to learn multidimensional representations of objects and their spatial relationships for purposes of image understanding. The What-and-Where filter is inspired by neurobiological data showing that a Where processing stream in the cerebral cortex is used for attentive spatial localization and orientation, whereas a What processing stream is used for attentive object learning and recognition.

A What-and-Where Neural Network for Invariant Image Preprocessing

A feedforward neural network for invariant image preprocessing is proposed that represents the position1 orientation and size of an image figure (where it is) in a multiplexed spatial map. This map is used to generate an invariant representation of the figure that is insensitive to position1 orientation, and size for purposes of pattern recognition (what it is). A multiscale array of oriented filters followed by competition between orientations and scales is used to define the Where filter.

Thomas Davis's education policies: theory and practice

This thesis explores the education policies of Thomas Davis. On the eve of the Great Famine Ireland was economically impoverished and politically dependent. The Irish people had a subservient mentality, were mainly uneducated and were unaware of their potential. He believed that education would develop a self-reliant, self-sufficient people; it would create a new generation of leaders and citizens necessary to transform Ireland into a prosperous, independent nation. This thesis explores his education philosophy which was political in orientation; he called for reform of university education so that it would educate leaders who were knowledgeable, patriotic and responsible. He formulated a curriculum which consisted of knowledge that would have direct use and application in public life; his curriculum included moral philosophy, oratory, philological studies and history. His contribution to the debate on the Queens Colleges bill, 1845, is explored including his public disagreement with Daniel O’Connell on the principle of multi-denominational education. This work also examines his policies on learning methodologies and teaching methods. It provides details of his thoughts on learning by experience, by observation, book learning and learning in the home. It focuses on the deficiencies evident in the system of teaching and learning that operated in Trinity College Dublin and it provides an analysis of his preferred method of instruction: Lyceum teaching. This thesis also explores his national curriculum in history and Irish culture which was designed to forge a sense of national identity, to win support for repeal and to develop the principle of nationality. He formulated a national curriculum to counteract the absence of national knowledge in the state schools, to provide the people with a positive self-image and ultimately to empower them to reclaim Ireland and to develop it. Davis knew the power of education and he used it as an instrument of political and social change.

Body image and body satisfaction differ by race in overweight postpartum mothers.

BACKGROUND: Body image (BI) and body satisfaction may be important in understanding weight loss behaviors, particularly during the postpartum period. We assessed these constructs among African American and white overweight postpartum women. METHODS: The sample included 162 women (73 African American and 89 white) in the intervention arm 6 months into the Active Mothers Postpartum (AMP) Study, a nutritional and physical activity weight loss intervention. BIs, self-reported using the Stunkard figure rating scale, were compared assessing mean values by race. Body satisfaction was measured using body discrepancy (BD), calculated as perceived current image minus ideal image (BD<0: desire to be heavier; BD>0: desire to be lighter). BD was assessed by race for: BD(Ideal) (current image minus the ideal image) and BD(Ideal Mother) (current image minus ideal mother image). RESULTS: Compared with white women, African American women were younger and were less likely to report being married, having any college education, or residing in households with annual incomes >$30,000 (all p < 0.01). They also had a higher mean body mass index (BMI) (p = 0.04), although perceived current BI did not differ by race (p = 0.21). African Americans had higher mean ideal (p = 0.07) and ideal mother (p = 0.001) BIs compared with whites. African Americans' mean BDs (adjusting for age, BMI, education, income, marital status, and interaction terms) were significantly lower than those of whites, indicating greater body satisfaction among African Americans (BD(Ideal): 1.7 vs. 2.3, p = 0.005; BD(Ideal Mother): 1.1 vs. 1.8, p = 0.0002). CONCLUSIONS: Racial differences exist in postpartum weight, ideal images, and body satisfaction. Healthcare providers should consider tailored messaging that accounts for these racially different perceptions and factors when designing weight loss programs for overweight mothers.

Making sense of health information technology implementation: A qualitative study protocol.

BACKGROUND: Implementing new practices, such as health information technology (HIT), is often difficult due to the disruption of the highly coordinated, interdependent processes (e.g., information exchange, communication, relationships) of providing care in hospitals. Thus, HIT implementation may occur slowly as staff members observe and make sense of unexpected disruptions in care. As a critical organizational function, sensemaking, defined as the social process of searching for answers and meaning which drive action, leads to unified understanding, learning, and effective problem solving -- strategies that studies have linked to successful change. Project teamwork is a change strategy increasingly used by hospitals that facilitates sensemaking by providing a formal mechanism for team members to share ideas, construct the meaning of events, and take next actions. METHODS: In this longitudinal case study, we aim to examine project teams' sensemaking and action as the team prepares to implement new information technology in a tiertiary care hospital. Based on management and healthcare literature on HIT implementation and project teamwork, we chose sensemaking as an alternative to traditional models for understanding organizational change and teamwork. Our methods choices are derived from this conceptual framework. Data on project team interactions will be prospectively collected through direct observation and organizational document review. Through qualitative methods, we will identify sensemaking patterns and explore variation in sensemaking across teams. Participant demographics will be used to explore variation in sensemaking patterns. DISCUSSION: Outcomes of this research will be new knowledge about sensemaking patterns of project teams, such as: the antecedents and consequences of the ongoing, evolutionary, social process of implementing HIT; the internal and external factors that influence the project team, including team composition, team member interaction, and interaction between the project team and the larger organization; the ways in which internal and external factors influence project team processes; and the ways in which project team processes facilitate team task accomplishment. These findings will lead to new methods of implementing HIT in hospitals.

Characterization of Image Quality for 3D Scatter Corrected Breast CT Images.

The goal of this study was to characterize the image quality of our dedicated, quasi-monochromatic spectrum, cone beam breast imaging system under scatter corrected and non-scatter corrected conditions for a variety of breast compositions. CT projections were acquired of a breast phantom containing two concentric sets of acrylic spheres that varied in size (1-8mm) based on their polar position. The breast phantom was filled with 3 different concentrations of methanol and water, simulating a range of breast densities (0.79-1.0g/cc); acrylic yarn was sometimes included to simulate connective tissue of a breast. For each phantom condition, 2D scatter was measured for all projection angles. Scatter-corrected and uncorrected projections were then reconstructed with an iterative ordered subsets convex algorithm. Reconstructed image quality was characterized using SNR and contrast analysis, and followed by a human observer detection task for the spheres in the different concentric rings. Results show that scatter correction effectively reduces the cupping artifact and improves image contrast and SNR. Results from the observer study indicate that there was no statistical difference in the number or sizes of lesions observed in the scatter versus non-scatter corrected images for all densities. Nonetheless, applying scatter correction for differing breast conditions improves overall image quality.

Exploring the effects of image persistence in low frame rate virtual environments

© 2015 IEEE.In virtual reality applications, there is an aim to provide real time graphics which run at high refresh rates. However, there are many situations in which this is not possible due to simulation or rendering issues. When running at low frame rates, several aspects of the user experience are affected. For example, each frame is displayed for an extended period of time, causing a high persistence image artifact. The effect of this artifact is that movement may lose continuity, and the image jumps from one frame to another. In this paper, we discuss our initial exploration of the effects of high persistence frames caused by low refresh rates and compare it to high frame rates and to a technique we developed to mitigate the effects of low frame rates. In this technique, the low frame rate simulation images are displayed with low persistence by blanking out the display during the extra time such image would be displayed. In order to isolate the visual effects, we constructed a simulator for low and high persistence displays that does not affect input latency. A controlled user study comparing the three conditions for the tasks of 3D selection and navigation was conducted. Results indicate that the low persistence display technique may not negatively impact user experience or performance as compared to the high persistence case. Directions for future work on the use of low persistence displays for low frame rate situations are discussed.

WormSizer: high-throughput analysis of nematode size and shape.

The fundamental phenotypes of growth rate, size and morphology are the result of complex interactions between genotype and environment. We developed a high-throughput software application, WormSizer, which computes size and shape of nematodes from brightfield images. Existing methods for estimating volume either coarsely model the nematode as a cylinder or assume the worm shape or opacity is invariant. Our estimate is more robust to changes in morphology or optical density as it only assumes radial symmetry. This open source software is written as a plugin for the well-known image-processing framework Fiji/ImageJ. It may therefore be extended easily. We evaluated the technical performance of this framework, and we used it to analyze growth and shape of several canonical Caenorhabditis elegans mutants in a developmental time series. We confirm quantitatively that a Dumpy (Dpy) mutant is short and fat and that a Long (Lon) mutant is long and thin. We show that daf-2 insulin-like receptor mutants are larger than wild-type upon hatching but grow slow, and WormSizer can distinguish dauer larvae from normal larvae. We also show that a Small (Sma) mutant is actually smaller than wild-type at all stages of larval development. WormSizer works with Uncoordinated (Unc) and Roller (Rol) mutants as well, indicating that it can be used with mutants despite behavioral phenotypes. We used our complete data set to perform a power analysis, giving users a sense of how many images are needed to detect different effect sizes. Our analysis confirms and extends on existing phenotypic characterization of well-characterized mutants, demonstrating the utility and robustness of WormSizer.