Efficient Image and Video Representations for Retrieval

Image (Video) retrieval is an interesting problem of retrieving images (videos) similar to the query. Images (Videos) are represented in an input (feature) space and similar images (videos) are obtained by finding nearest neighbors in the input representation space. Numerous input representations both in real valued and binary space have been proposed for conducting faster retrieval. In this thesis, we present techniques that obtain improved input representations for retrieval in both supervised and unsupervised settings for images and videos. Supervised retrieval is a well known problem of retrieving same class images of the query. We address the practical aspects of achieving faster retrieval with binary codes as input representations for the supervised setting in the first part, where binary codes are used as addresses into hash tables. In practice, using binary codes as addresses does not guarantee fast retrieval, as similar images are not mapped to the same binary code (address). We address this problem by presenting an efficient supervised hashing (binary encoding) method that aims to explicitly map all the images of the same class ideally to a unique binary code. We refer to the binary codes of the images as `Semantic Binary Codes' and the unique code for all same class images as `Class Binary Code'. We also propose a new class­ based Hamming metric that dramatically reduces the retrieval times for larger databases, where only hamming distance is computed to the class binary codes. We also propose a Deep semantic binary code model, by replacing the output layer of a popular convolutional Neural Network (AlexNet) with the class binary codes and show that the hashing functions learned in this way outperforms the state­ of ­the art, and at the same time provide fast retrieval times. In the second part, we also address the problem of supervised retrieval by taking into account the relationship between classes. For a given query image, we want to retrieve images that preserve the relative order i.e. we want to retrieve all same class images first and then, the related classes images before different class images. We learn such relationship aware binary codes by minimizing the similarity between inner product of the binary codes and the similarity between the classes. We calculate the similarity between classes using output embedding vectors, which are vector representations of classes. Our method deviates from the other supervised binary encoding schemes as it is the first to use output embeddings for learning hashing functions. We also introduce new performance metrics that take into account the related class retrieval results and show significant gains over the state­ of­ the art. High Dimensional descriptors like Fisher Vectors or Vector of Locally Aggregated Descriptors have shown to improve the performance of many computer vision applications including retrieval. In the third part, we will discuss an unsupervised technique for compressing high dimensional vectors into high dimensional binary codes, to reduce storage complexity. In this approach, we deviate from adopting traditional hyperplane hashing functions and instead learn hyperspherical hashing functions. The proposed method overcomes the computational challenges of directly applying the spherical hashing algorithm that is intractable for compressing high dimensional vectors. A practical hierarchical model that utilizes divide and conquer techniques using the Random Select and Adjust (RSA) procedure to compress such high dimensional vectors is presented. We show that our proposed high dimensional binary codes outperform the binary codes obtained using traditional hyperplane methods for higher compression ratios. In the last part of the thesis, we propose a retrieval based solution to the Zero shot event classification problem - a setting where no training videos are available for the event. To do this, we learn a generic set of concept detectors and represent both videos and query events in the concept space. We then compute similarity between the query event and the video in the concept space and videos similar to the query event are classified as the videos belonging to the event. We show that we significantly boost the performance using concept features from other modalities.

Robust image hashing using higher order spectral features

Robust image hashing seeks to transform a given input image into a shorter hashed version using a key-dependent non-invertible transform. These image hashes can be used for watermarking, image integrity authentication or image indexing for fast retrieval. This paper introduces a new method of generating image hashes based on extracting Higher Order Spectral features from the Radon projection of an input image. The feature extraction process is non-invertible, non-linear and different hashes can be produced from the same image through the use of random permutations of the input. We show that the transform is robust to typical image transformations such as JPEG compression, noise, scaling, rotation, smoothing and cropping. We evaluate our system using a verification-style framework based on calculating false match, false non-match likelihoods using the publicly available Uncompressed Colour Image database (UCID) of 1320 images. We also compare our results to Swaminathan’s Fourier-Mellin based hashing method with at least 1% EER improvement under noise, scaling and sharpening.

On the (in)security of IDEA in various hashing modes

In this article, we study the security of the IDEA block cipher when it is used in various simple-length or double-length hashing modes. Even though this cipher is still considered as secure, we show that one should avoid its use as internal primitive for block cipher based hashing. In particular, we are able to generate instantaneously free-start collisions for most modes, and even semi-free-start collisions, pseudo-preimages or hash collisions in practical complexity. This work shows a practical example of the gap that exists between secret-key and known or chosen-key security for block ciphers. Moreover, we also settle the 20-year-old standing open question concerning the security of the Abreast-DM and Tandem-DM double-length compression functions, originally invented to be instantiated with IDEA. Our attacks have been verified experimentally and work even for strengthened versions of IDEA with any number of rounds.

Sibling bullying perpetration: Associations with gender, grade, peer perpetration, trait anger, and moral disengagement

This study investigated bullying amongst siblings in both traditional and cyber forms, and the associations of gender, grade, peer bullying perpetration, trait anger and moral disengagement. The participants were 455 children in grades 5 to 12 (262 girls and 177 boys with 16 unknown gender) who had a sibling. As the number of siblings who only bullied by technology was low, these associations were not able to be calculated. However, the findings showed that the percentage of sibling traditional bullying perpetration (31.6%) was higher than peer bullying perpetration (9.8%). Sibling bullies reported engaging in complex behaviours of perpetration and victimisation in both the physical and in cyber settings, although the number was small. Gender, trait anger, moral disengagement and bullying peers at school (but not grade) were all significantly associated with sibling traditional bullying perpetration. The implications of the findings are discussed for bullying intervention and prevention programs to understand childhood bullying in diverse contexts.

Going through the obstruction, the Torrens System Assurance Fund and contemporary solutions : a tale weaved from a story of a nun, a romantic triangle and sibling corruption

Private title insurance has been the subject of much debate by law reform bodies and academics. This article adds a new dimension to the discussion by analysing its role against a recent scenario where a nun was betrayed by the actions of her brother, and compensation payable from the assurance fund, after much challenge by the registrar, amounted to in excess of $4 million.We ask whether the slow burning of title insurance into the psyche of Australian home purchasers will see state-based assurance fundings looking to minismise their role in the Torrens system. We also query how the rather more immediate electronic establishment of electronic conveyancing will alter the balance between the assurance fund, private title insurance and the increasing responsibilities on stakeholdes involved in conveyancing.

Locality-sensitive hashing for protein classification

Determination of sequence similarity is a central issue in computational biology, a problem addressed primarily through BLAST, an alignment based heuristic which has underpinned much of the analysis and annotation of the genomic era. Despite their success, alignment-based approaches scale poorly with increasing data set size, and are not robust under structural sequence rearrangements. Successive waves of innovation in sequencing technologies – so-called Next Generation Sequencing (NGS) approaches – have led to an explosion in data availability, challenging existing methods and motivating novel approaches to sequence representation and similarity scoring, including adaptation of existing methods from other domains such as information retrieval. In this work, we investigate locality-sensitive hashing of sequences through binary document signatures, applying the method to a bacterial protein classification task. Here, the goal is to predict the gene family to which a given query protein belongs. Experiments carried out on a pair of small but biologically realistic datasets (the full protein repertoires of families of Chlamydia and Staphylococcus aureus genomes respectively) show that a measure of similarity obtained by locality sensitive hashing gives highly accurate results while offering a number of avenues which will lead to substantial performance improvements over BLAST..

Take my brother first : heroic final act saves sibling

"A schoolboy who begged rescuers to save his little brother first is among a dozen people drowned."

Morphology, echolocation and foraging behaviour in two sympatric sibling species of bat (Tylonycteris pachypus and Tylonycteris robustula) (Chiroptera : Vespertilionidae)

We studied the wing morphology, echolocation calls, foraging behaviour and flight speed of Tylonycteris pachypus and Tylonycteris robustula in Longzhou County, South China during the summer (June–August) of 2005. The wingspan, wing loading and aspect ratio of the two species were relatively low, and those of T. pachypus were lower compared with T. robustula. The echolocation calls of T. pachypus and T. robustula consist of a broadband frequency modulated (FM) sweep followed by a short narrowband FM sweep. The dominant frequency of calls of T. pachypus was 65.1 kHz, whereas that of T. robustula was 57.7 kHz. The call frequencies (including highest frequency of the call, lowest frequency of the call and frequency of the call that contained most energy) of T. pachypus were higher than those of T. robustula, and the pulse duration of the former was longer than that of the latter. The inter-pulse interval and bandwidth of the calls were not significantly different between the two species. Tylonycteris pachypus foraged in more complex environments than T. robustula, although the two species were both netted in edge habitats (around trees or houses), along pathways and in the tops of trees. Tylonycteris pachypus flew slower (straight level flight speed, 4.3 m s−1) than T. robustula (straight level flight speed, 4.8 m s−1). We discuss the relationship between wing morphology, echolocation calls, foraging behaviour and flight speed, and demonstrate resource partitioning between these two species in terms of morphological and behavioural factors.

Do twins differ from single-born children on rates of behavioral difficulty in early childhood? A study of sibling relationship risk factors

The hypothesis that twinning raises risk for behavioral difficulties in childhood is persistent, yet there is limited and inconsistent empirical evidence. Simple mean comparison without control for confounders provides data on prevalence rates but cannot provide knowledge about risk or etiology. To assess the effect of twin relationship on behavior, comparison of patterns of association with single-born siblings may be informative. Analyses of data from an Australian sample of twins and single-born children (N = 305, mean age 4 years 9 months, and a follow-up 12 months later) were undertaken. The outcome measure was the Strengths and Difficulties Questionnaire. Predictor and control measures were obtained from parent report on the sibling/co-twin relationship behavior, family demographics, and obstetric history. We assessed difference between twins and single-born children in two respects: (a) mean behavioral difficulties, and (b) patterns of association between sibling relationship and behavioral difficulties, controlling for confounders. Results showed no differences in mean levels of behavioral difficulties between twins and single-born siblings identifying the importance of statistical control for family and obstetric adversity. Differences in patterns of association were found; for twin children, conflict in their co-twin relationship predicted externalizing behaviors, while for single-born children conflict predicted internalizing behaviors. The findings of mean differences between twin and single-born children in social background, but not in behavioral difficulties, underscore the necessity of statistical control to identify risk associated with twinning compared with risk associated with family and obstetric background factors.

Mothers' perception of the quality of childhood sibling relationships affected by disability

The quality of the sibling relationship has an important role in the development of psychosocial skills throughout childhood. While the literature suggests that the significance of sibling relationships is heightened when one sibling has a disability, empirical findings about the quality of these relationships are few and inconsistent. The present study aimed to address this gap, by investigating mothers’ perspectives about the impact of disability on the quality of the childhood sibling relationship. Forty-one mothers with a child with disability, and 48 with no children with disability completed an online questionnaire that assessed the amount of perceived warmth/closeness and conflict in their children’s sibling relationship. It was found that while there were no differences in reported conflict between the two groups, mothers with a child with disability reported significantly lower warmth/closeness in their children’s sibling relationship than mothers without a child with disability. Demographic variables such as number of children, gender grouping, target gender, target age and age order did not moderate this result. Mothers overall reported significantly more warmth/closeness for younger rather than older children, and more conflict when the sibling was younger than the target child as opposed to older than them. Clinical implications for intervention are discussed.

Whole-genome linkage analysis of rheumatoid arthritis susceptibility loci in 252 affected sibling pairs in the United Kingdom

Objective. To undertake a systematic wholegenome screen to identify regions exhibiting genetic linkage to rheumatoid arthritis (RA). Methods. Two hundred fifty-two RA-affected sibling pairs from 182 UK families were genotyped using 365 highly informative microsatellite markers. Microsatellite genotyping was performed using fluorescent polymerase chain reaction primers and semiautomated DNA sequencing technology. Linkage analysis was undertaken using MAPMAKER/SIBS for single-point and multipoint analysis. Results. Significant linkage (maximum logarithm of odds score 4.7 [P = 0.000003] at marker D6S276, 1 cM from HLA-DRB1) was identified around the major histocompatibility complex (MHC) region on chromosome 6. Suggestive linkage (P < 7.4 × 10-4) was identified on chromosome 6q by single- and multipoint analysis. Ten other sites of nominal linkage (P < 0.05) were identified on chromosomes 3p, 4q, 7p, 2 regions of 10q, 2 regions of 14q, 16p, 21q, and Xq by single-point analysis and on 3 sites (1q, 14q, and 14q) by multipoint analysis. Conclusion. Linkage to the MHC region was confirmed. Eleven non-HLA regions demonstrated evidence of suggestive or nominal linkage, but none reached the genome-wide threshold for significant linkage (P = 2.2 × 10-5). Results of previous genome screens have suggested that 6 of these regions may be involved in RA susceptibility.

Scalable document hashing and retrieval

This thesis studies document signatures, which are small representations of documents and other objects that can be stored compactly and compared for similarity. This research finds that document signatures can be effectively and efficiently used to both search and understand relationships between documents in large collections, scalable enough to search a billion documents in a fraction of a second. Deliverables arising from the research include an investigation of the representational capacity of document signatures, the publication of an open-source signature search platform and an approach for scaling signature retrieval to operate efficiently on collections containing hundreds of millions of documents.

Inference of the genetic architecture underlying BMI and height with the use of 20,240 sibling pairs

Evidence that complex traits are highly polygenic has been presented by population-based genome-wide association studies (GWASs) through the identification of many significant variants, as well as by family-based de novo sequencing studies indicating that several traits have a large mutational target size. Here, using a third study design, we show results consistent with extreme polygenicity for body mass index (BMI) and height. On a sample of 20,240 siblings (from 9,570 nuclear families), we used a within-family method to obtain narrow-sense heritability estimates of 0.42 (SE = 0.17, p = 0.01) and 0.69 (SE = 0.14, p = 6 x 10(-)(7)) for BMI and height, respectively, after adjusting for covariates. The genomic inflation factors from locus-specific linkage analysis were 1.69 (SE = 0.21, p = 0.04) for BMI and 2.18 (SE = 0.21, p = 2 x 10(-10)) for height. This inflation is free of confounding and congruent with polygenicity, consistent with observations of ever-increasing genomic-inflation factors from GWASs with large sample sizes, implying that those signals are due to true genetic signals across the genome rather than population stratification. We also demonstrate that the distribution of the observed test statistics is consistent with both rare and common variants underlying a polygenic architecture and that previous reports of linkage signals in complex traits are probably a consequence of polygenic architecture rather than the segregation of variants with large effects. The convergent empirical evidence from GWASs, de novo studies, and within-family segregation implies that family-based sequencing studies for complex traits require very large sample sizes because the effects of causal variants are small on average.

Within-family outliers: Segregating alleles or environmental effects? A linkage analysis of height from 5815 sibling pairs

Most information in linkage analysis for quantitative traits comes from pairs of relatives that are phenotypically most discordant or concordant. Confounding this, within-family outliers from non-genetic causes may create false positives and negatives. We investigated the influence of within-family outliers empirically, using one of the largest genome-wide linkage scans for height. The subjects were drawn from Australian twin cohorts consisting of 8447 individuals in 2861 families, providing a total of 5815 possible pairs of siblings in sibships. A variance component linkage analysis was performed, either including or excluding the within-family outliers. Using the entire dataset, the largest LOD scores were on chromosome 15q (LOD 2.3) and 11q (1.5). Excluding within-family outliers increased the LOD score for most regions, but the LOD score on chromosome 15 decreased from 2.3 to 1.2, suggesting that the outliers may create false negatives and false positives, although rare alleles of large effect may also be an explanation. Several regions suggestive of linkage to height were found after removing the outliers, including 1q23.1 (2.0), 3q22.1 (1.9) and 5q32 (2.3). We conclude that the investigation of the effect of within-family outliers, which is usually neglected, should be a standard quality control measure in linkage analysis for complex traits and may reduce the noise for the search of common variants of modest effect size as well as help identify rare variants of large effect and clinical significance. We suggest that the effect of within-family outliers deserves further investigation via theoretical and simulation studies.