Learning Binary Code Representations for Effective and Efficient Image Retrieval

The size of online image datasets is constantly increasing. Considering an image dataset with millions of images, image retrieval becomes a seemingly intractable problem for exhaustive similarity search algorithms. Hashing methods, which encodes high-dimensional descriptors into compact binary strings, have become very popular because of their high efficiency in search and storage capacity. In the first part, we propose a multimodal retrieval method based on latent feature models. The procedure consists of a nonparametric Bayesian framework for learning underlying semantically meaningful abstract features in a multimodal dataset, a probabilistic retrieval model that allows cross-modal queries and an extension model for relevance feedback. In the second part, we focus on supervised hashing with kernels. We describe a flexible hashing procedure that treats binary codes and pairwise semantic similarity as latent and observed variables, respectively, in a probabilistic model based on Gaussian processes for binary classification. We present a scalable inference algorithm with the sparse pseudo-input Gaussian process (SPGP) model and distributed computing. In the last part, we define an incremental hashing strategy for dynamic databases where new images are added to the databases frequently. The method is based on a two-stage classification framework using binary and multi-class SVMs. The proposed method also enforces balance in binary codes by an imbalance penalty to obtain higher quality binary codes. We learn hash functions by an efficient algorithm where the NP-hard problem of finding optimal binary codes is solved via cyclic coordinate descent and SVMs are trained in a parallelized incremental manner. For modifications like adding images from an unseen class, we propose an incremental procedure for effective and efficient updates to the previous hash functions. Experiments on three large-scale image datasets demonstrate that the incremental strategy is capable of efficiently updating hash functions to the same retrieval performance as hashing from scratch.

TRANSCENDING FACH: A SEARCH FOR IDENTITY INSIDE AND OUT OF MEZZO-SOPRANO REPERTOIRE

The German Fach system is a tool to classify voices in classical singing. This dissertation comprises three different programs that reflect my search for identity as a mezzo-soprano and my desire to transcend the limitations of Fach. The three programs, all presented at The Clarice Performing Arts Center, contain repertoire written for male and female voices thus allowing me to explore areas outside of the mezzo-soprano Fach, gain a better understanding of the Fach system and guide me as I strive to become a more mature performer. In my first program, I sang the role of Sesto, a role that was composed originally for a castrate, in the opera La Clemenza di Tito by W.A. Mozart. The Maryland Opera Studio production took place April 30, May 2,4&6,2003. Performing this gender-bending role provided an experience of physical behavior from the male view point along with the demands of coloratura singing. Program two (November 30,2004) contained the song cycle Dichterliebe by Robert Schumann and songs by Ludwig van Beethoven, Franz Schubert and Felix Mendelssohn, which are usually sung by male voices. This program experimented with extended range, tessitura and a gender-bending performance in the art song arena. 8 In program three (April 21 &23,2005), I sang the contralto role of Cornelia from Giulio Cesare in Egitto by George Frederic Handel. The role of Cornelia is psychologically complex, expressing emotions such as love, melancholy, rage, malice, joy and fear. To convey these emotions a voice needs warmth and darkness of quality. Although the range is close to that of the mezzo-soprano, Handel wrote Cornelia for contralto voice because he wanted a dark timbre and this role allowed me to develop my lower register and manage suitable ornamentations. The programs are documented in a digital format available on compact disc and are accompanied by the oral presentation at the defense of this dissertation.

IN SEARCH OF THE DIVINE: SPIRITUALITY AND PRAYER IN DIVERSE VIOLIN REPERTOIRE FROM BIBER TO IVES

Composers from all eras and of all ethnicities explore spirituality and prayer by using one or a combination of the following ideas: having a spiritual concept in mind when composing certain pieces, quoting hymns, being influenced by their own personal beliefs, or portraying spiritual figures and ideas in their works. Some musical works are inspired by spirituality; others, as in the case of Bloch's Nigun, even serve as prayers themselves. These recitals gave me the opportunity to approach a wide variety of musical styles while discovering my own mode for expression. The unaccompanied violin works throughout this project trace a distinct lineage from the baroque to the twentieth century. Biber's appendix to the Rosary Sonatas, the Passacaglia for solo violin, is a crucial predecessor to Bach's monumental Chaconne. Eugene Ysaye was inspired to write the Six Sonatas, Op. 27 after he attended a performance of Bach's Sonatas and Partitas given by Josef Szigeti. Ysaye's second solo sonata blatantly quotes Bach's Partita No.3 in E major throughout the first movement. Every movement also contains quotations from and variations on the plainchant Dies Irae. Although each of the solo violin works presented in this project may be viewed as virtuosic concert pieces, each piece allows the performer to transcend the technical hurdles-and perhaps even utilize them-to serve a higher, artistic and spiritual purpose while alone on the concert stage. Each of the sonata works in this project requires a close, equal collaboration between violinist and pianist, rather than displaying the violinist as soloist.