BoostMap: A Method for Efficient Approximate Similarity Rankings

This paper introduces BoostMap, a method that can significantly reduce retrieval time in image and video database systems that employ computationally expensive distance measures, metric or non-metric. Database and query objects are embedded into a Euclidean space, in which similarities can be rapidly measured using a weighted Manhattan distance. Embedding construction is formulated as a machine learning task, where AdaBoost is used to combine many simple, 1D embeddings into a multidimensional embedding that preserves a significant amount of the proximity structure in the original space. Performance is evaluated in a hand pose estimation system, and a dynamic gesture recognition system, where the proposed method is used to retrieve approximate nearest neighbors under expensive image and video similarity measures. In both systems, BoostMap significantly increases efficiency, with minimal losses in accuracy. Moreover, the experiments indicate that BoostMap compares favorably with existing embedding methods that have been employed in computer vision and database applications, i.e., FastMap and Bourgain embeddings.

Towards effective indexing for very large video sequence database

With rapid advances in video processing technologies and ever fast increments in network bandwidth, the popularity of video content publishing and sharing has made similarity search an indispensable operation to retrieve videos of user interests. The video similarity is usually measured by the percentage of similar frames shared by two video sequences, and each frame is typically represented as a high-dimensional feature vector. Unfortunately, high complexity of video content has posed the following major challenges for fast retrieval: (a) effective and compact video representations, (b) efficient similarity measurements, and (c) efficient indexing on the compact representations. In this paper, we propose a number of methods to achieve fast similarity search for very large video database. First, each video sequence is summarized into a small number of clusters, each of which contains similar frames and is represented by a novel compact model called Video Triplet (ViTri). ViTri models a cluster as a tightly bounded hypersphere described by its position, radius, and density. The ViTri similarity is measured by the volume of intersection between two hyperspheres multiplying the minimal density, i.e., the estimated number of similar frames shared by two clusters. The total number of similar frames is then estimated to derive the overall similarity between two video sequences. Hence the time complexity of video similarity measure can be reduced greatly. To further reduce the number of similarity computations on ViTris, we introduce a new one dimensional transformation technique which rotates and shifts the original axis system using PCA in such a way that the original inter-distance between two high-dimensional vectors can be maximally retained after mapping. An efficient B+-tree is then built on the transformed one dimensional values of ViTris' positions. Such a transformation enables B+-tree to achieve its optimal performance by quickly filtering a large portion of non-similar ViTris. Our extensive experiments on real large video datasets prove the effectiveness of our proposals that outperform existing methods significantly.

An alignment based similarity measure for hand detection in cluttered sign language video

Locating hands in sign language video is challenging due to a number of factors. Hand appearance varies widely across signers due to anthropometric variations and varying levels of signer proficiency. Video can be captured under varying illumination, camera resolutions, and levels of scene clutter, e.g., high-res video captured in a studio vs. low-res video gathered by a web cam in a user’s home. Moreover, the signers’ clothing varies, e.g., skin-toned clothing vs. contrasting clothing, short-sleeved vs. long-sleeved shirts, etc. In this work, the hand detection problem is addressed in an appearance matching framework. The Histogram of Oriented Gradient (HOG) based matching score function is reformulated to allow non-rigid alignment between pairs of images to account for hand shape variation. The resulting alignment score is used within a Support Vector Machine hand/not-hand classifier for hand detection. The new matching score function yields improved performance (in ROC area and hand detection rate) over the Vocabulary Guided Pyramid Match Kernel (VGPMK) and the traditional, rigid HOG distance on American Sign Language video gestured by expert signers. The proposed match score function is computationally less expensive (for training and testing), has fewer parameters and is less sensitive to parameter settings than VGPMK. The proposed detector works well on test sequences from an inexpert signer in a non-studio setting with cluttered background.

Hierarchical indexing structure for efficient similarity search in video retrieval

With the rapid increase in both centralized video archives and distributed WWW video resources, content-based video retrieval is gaining its importance. To support such applications efficiently, content-based video indexing must be addressed. Typically, each video is represented by a sequence of frames. Due to the high dimensionality of frame representation and the large number of frames, video indexing introduces an additional degree of complexity. In this paper, we address the problem of content-based video indexing and propose an efficient solution, called the Ordered VA-File (OVA-File) based on the VA-file. OVA-File is a hierarchical structure and has two novel features: 1) partitioning the whole file into slices such that only a small number of slices are accessed and checked during k Nearest Neighbor (kNN) search and 2) efficient handling of insertions of new vectors into the OVA-File, such that the average distance between the new vectors and those approximations near that position is minimized. To facilitate a search, we present an efficient approximate kNN algorithm named Ordered VA-LOW (OVA-LOW) based on the proposed OVA-File. OVA-LOW first chooses possible OVA-Slices by ranking the distances between their corresponding centers and the query vector, and then visits all approximations in the selected OVA-Slices to work out approximate kNN. The number of possible OVA-Slices is controlled by a user-defined parameter delta. By adjusting delta, OVA-LOW provides a trade-off between the query cost and the result quality. Query by video clip consisting of multiple frames is also discussed. Extensive experimental studies using real video data sets were conducted and the results showed that our methods can yield a significant speed-up over an existing VA-file-based method and iDistance with high query result quality. Furthermore, by incorporating temporal correlation of video content, our methods achieved much more efficient performance.

Automatically detecting pain in video through facial action units

In a clinical setting, pain is reported either through patient self-report or via an observer. Such measures are problematic as they are: 1) subjective, and 2) give no specific timing information. Coding pain as a series of facial action units (AUs) can avoid these issues as it can be used to gain an objective measure of pain on a frame-by-frame basis. Using video data from patients with shoulder injuries, in this paper, we describe an active appearance model (AAM)-based system that can automatically detect the frames in video in which a patient is in pain. This pain data set highlights the many challenges associated with spontaneous emotion detection, particularly that of expression and head movement due to the patient's reaction to pain. In this paper, we show that the AAM can deal with these movements and can achieve significant improvements in both the AU and pain detection performance compared to the current-state-of-the-art approaches which utilize similarity-normalized appearance features only.

SeamSeg: Video Object Segmentation using Patch Seams

In this paper, we propose a technique for video object segmentation using patch seams across frames. Typically, seams, which are connected paths of low energy, are utilised for retargeting, where the primary aim is to reduce the image size while preserving the salient image contents. Here, we adapt the formulation of seams for temporal label propagation. The energy function associated with the proposed video seams provides temporal linking of patches across frames, to accurately segment the object. The proposed energy function takes into account the similarity of patches along the seam, temporal consistency of motion and spatial coherency of seams. Label propagation is achieved with high fidelity in the critical boundary regions, utilising the proposed patch seams. To achieve this without additional overheads, we curtail the error propagation by formulating boundary regions as rough-sets. The proposed approach out-perform state-of-the-art supervised and unsupervised algorithms, on benchmark datasets.

An Invariant Representation for Matching Trajectories across Uncalibrated Video Streams

We introduce a view-point invariant representation of moving object trajectories that can be used in video database applications. It is assumed that trajectories lie on a surface that can be locally approximated with a plane. Raw trajectory data is first locally approximated with a cubic spline via least squares fitting. For each sampled point of the obtained curve, a projective invariant feature is computed using a small number of points in its neighborhood. The resulting sequence of invariant features computed along the entire trajectory forms the view invariant descriptor of the trajectory itself. Time parametrization has been exploited to compute cross ratios without ambiguity due to point ordering. Similarity between descriptors of different trajectories is measured with a distance that takes into account the statistical properties of the cross ratio, and its symmetry with respect to the point at infinity. In experiments, an overall correct classification rate of about 95% has been obtained on a dataset of 58 trajectories of players in soccer video, and an overall correct classification rate of about 80% has been obtained on matching partial segments of trajectories collected from two overlapping views of outdoor scenes with moving people and cars.

Hybrid Background Subtraction in Video using Bi-level CodeBook Model

Detection of Objects in Video is a highly demanding area of research. The Background Subtraction Algorithms can yield better results in Foreground Object Detection. This work presents a Hybrid CodeBook based Background Subtraction to extract the foreground ROI from the background. Codebooks are used to store compressed information by demanding lesser memory usage and high speedy processing. This Hybrid method which uses Block-Based and Pixel-Based Codebooks provide efficient detection results; the high speed processing capability of block based background subtraction as well as high Precision Rate of pixel based background subtraction are exploited to yield an efficient Background Subtraction System. The Block stage produces a coarse foreground area, which is then refined by the Pixel stage. The system’s performance is evaluated with different block sizes and with different block descriptors like 2D-DCT, FFT etc. The Experimental analysis based on statistical measurements yields precision, recall, similarity and F measure of the hybrid system as 88.74%, 91.09%, 81.66% and 89.90% respectively, and thus proves the efficiency of the novel system.

"Thematically Analysing Social Network Content During Disasters Through the Lens of the Disaster Management Lifecycle" & "Investigating Similarity Between Privacy Policies of Social Networking Sites as a Precursor for Standardization"

Abstract 1: Social Networks such as Twitter are often used for disseminating and collecting information during natural disasters. The potential for its use in Disaster Management has been acknowledged. However, more nuanced understanding of the communications that take place on social networks are required to more effectively integrate this information into the processes within disaster management. The type and value of information shared should be assessed, determining the benefits and issues, with credibility and reliability as known concerns. Mapping the tweets in relation to the modelled stages of a disaster can be a useful evaluation for determining the benefits/drawbacks of using data from social networks, such as Twitter, in disaster management.A thematic analysis of tweets’ content, language and tone during the UK Storms and Floods 2013/14 was conducted. Manual scripting was used to determine the official sequence of events, and classify the stages of the disaster into the phases of the Disaster Management Lifecycle, to produce a timeline. Twenty- five topics discussed on Twitter emerged, and three key types of tweets, based on the language and tone, were identified. The timeline represents the events of the disaster, according to the Met Office reports, classed into B. Faulkner’s Disaster Management Lifecycle framework. Context is provided when observing the analysed tweets against the timeline. This illustrates a potential basis and benefit for mapping tweets into the Disaster Management Lifecycle phases. Comparing the number of tweets submitted in each month with the timeline, suggests users tweet more as an event heightens and persists. Furthermore, users generally express greater emotion and urgency in their tweets.This paper concludes that the thematic analysis of content on social networks, such as Twitter, can be useful in gaining additional perspectives for disaster management. It demonstrates that mapping tweets into the phases of a Disaster Management Lifecycle model can have benefits in the recovery phase, not just in the response phase, to potentially improve future policies and activities. Abstract2: The current execution of privacy policies, as a mode of communicating information to users, is unsatisfactory. Social networking sites (SNS) exemplify this issue, attracting growing concerns regarding their use of personal data and its effect on user privacy. This demonstrates the need for more informative policies. However, SNS lack the incentives required to improve policies, which is exacerbated by the difficulties of creating a policy that is both concise and compliant. Standardization addresses many of these issues, providing benefits for users and SNS, although it is only possible if policies share attributes which can be standardized. This investigation used thematic analysis and cross- document structure theory, to assess the similarity of attributes between the privacy policies (as available in August 2014), of the six most frequently visited SNS globally. Using the Jaccard similarity coefficient, two types of attribute were measured; the clauses used by SNS and the coverage of forty recommendations made by the UK Information Commissioner’s Office. Analysis showed that whilst similarity in the clauses used was low, similarity in the recommendations covered was high, indicating that SNS use different clauses, but to convey similar information. The analysis also showed that low similarity in the clauses was largely due to differences in semantics, elaboration and functionality between SNS. Therefore, this paper proposes that the policies of SNS already share attributes, indicating the feasibility of standardization and five recommendations are made to begin facilitating this, based on the findings of the investigation.

High definition video quality assessment metric built upon full reference ratios

Métrica de calidad de video de alta definición construida a partir de ratios de referencia completa. La medida de calidad de video, en inglés Visual Quality Assessment (VQA), es uno de los mayores retos por solucionar en el entorno multimedia. La calidad de vídeo tiene un impacto altísimo en la percepción del usuario final (consumidor) de los servicios sustentados en la provisión de contenidos multimedia y, por tanto, factor clave en la valoración del nuevo paradigma denominado Calidad de la Experiencia, en inglés Quality of Experience (QoE). Los modelos de medida de calidad de vídeo se pueden agrupar en varias ramas según la base técnica que sustenta el sistema de medida, destacando en importancia los que emplean modelos psicovisuales orientados a reproducir las características del sistema visual humano, en inglés Human Visual System, del que toman sus siglas HVS, y los que, por el contrario, optan por una aproximación ingenieril en la que el cálculo de calidad está basado en la extracción de parámetros intrínsecos de la imagen y su comparación. A pesar de los avances recogidos en este campo en los últimos años, la investigación en métricas de calidad de vídeo, tanto en presencia de referencia (los modelos denominados de referencia completa), como en presencia de parte de ella (modelos de referencia reducida) e incluso los que trabajan en ausencia de la misma (denominados sin referencia), tiene un amplio camino de mejora y objetivos por alcanzar. Dentro de ellos, la medida de señales de alta definición, especialmente las utilizadas en las primeras etapas de la cadena de valor que son de muy alta calidad, son de especial interés por su influencia en la calidad final del servicio y no existen modelos fiables de medida en la actualidad. Esta tesis doctoral presenta un modelo de medida de calidad de referencia completa que hemos llamado PARMENIA (PArallel Ratios MEtric from iNtrInsic features Analysis), basado en la ponderación de cuatro ratios de calidad calculados a partir de características intrínsecas de la imagen. Son: El Ratio de Fidelidad, calculado mediante el gradiente morfológico o gradiente de Beucher. El Ratio de Similitud Visual, calculado mediante los puntos visualmente significativos de la imagen a través de filtrados locales de contraste. El Ratio de Nitidez, que procede de la extracción del estadístico de textura de Haralick contraste. El Ratio de Complejidad, obtenido de la definición de homogeneidad del conjunto de estadísticos de textura de Haralick PARMENIA presenta como novedad la utilización de la morfología matemática y estadísticos de Haralick como base de una métrica de medida de calidad, pues esas técnicas han estado tradicionalmente más ligadas a la teledetección y la segmentación de objetos. Además, la aproximación de la métrica como un conjunto ponderado de ratios es igualmente novedosa debido a que se alimenta de modelos de similitud estructural y otros más clásicos, basados en la perceptibilidad del error generado por la degradación de la señal asociada a la compresión. PARMENIA presenta resultados con una altísima correlación con las valoraciones MOS procedentes de las pruebas subjetivas a usuarios que se han realizado para la validación de la misma. El corpus de trabajo seleccionado procede de conjuntos de secuencias validados internacionalmente, de modo que los resultados aportados sean de la máxima calidad y el máximo rigor posible. La metodología de trabajo seguida ha consistido en la generación de un conjunto de secuencias de prueba de distintas calidades a través de la codificación con distintos escalones de cuantificación, la obtención de las valoraciones subjetivas de las mismas a través de pruebas subjetivas de calidad (basadas en la recomendación de la Unión Internacional de Telecomunicaciones BT.500), y la validación mediante el cálculo de la correlación de PARMENIA con estos valores subjetivos, cuantificada a través del coeficiente de correlación de Pearson. Una vez realizada la validación de los ratios y optimizada su influencia en la medida final y su alta correlación con la percepción, se ha realizado una segunda revisión sobre secuencias del hdtv test dataset 1 del Grupo de Expertos de Calidad de Vídeo (VQEG, Video Quality Expert Group) mostrando los resultados obtenidos sus claras ventajas. Abstract Visual Quality Assessment has been so far one of the most intriguing challenges on the media environment. Progressive evolution towards higher resolutions while increasing the quality needed (e.g. high definition and better image quality) aims to redefine models for quality measuring. Given the growing interest in multimedia services delivery, perceptual quality measurement has become a very active area of research. First, in this work, a classification of objective video quality metrics based on their underlying methodologies and approaches for measuring video quality has been introduced to sum up the state of the art. Then, this doctoral thesis describes an enhanced solution for full reference objective quality measurement based on mathematical morphology, texture features and visual similarity information that provides a normalized metric that we have called PARMENIA (PArallel Ratios MEtric from iNtrInsic features Analysis), with a high correlated MOS score. The PARMENIA metric is based on the pooling of different quality ratios that are obtained from three different approaches: Beucher’s gradient, local contrast filtering, and contrast and homogeneity Haralick’s texture features. The metric performance is excellent, and improves the current state of the art by providing a wide dynamic range that make easier to discriminate between very close quality coded sequences, especially for very high bit rates whose quality, currently, is transparent for quality metrics. PARMENIA introduces a degree of novelty against other working metrics: on the one hand, exploits the structural information variation to build the metric’s kernel, but complements the measure with texture information and a ratio of visual meaningful points that is closer to typical error sensitivity based approaches. We would like to point out that PARMENIA approach is the only metric built upon full reference ratios, and using mathematical morphology and texture features (typically used in segmentation) for quality assessment. On the other hand, it gets results with a wide dynamic range that allows measuring the quality of high definition sequences from bit rates of hundreds of Megabits (Mbps) down to typical distribution rates (5-6 Mbps), even streaming rates (1- 2 Mbps). Thus, a direct correlation between PARMENIA and MOS scores are easily constructed. PARMENIA may further enhance the number of available choices in objective quality measurement, especially for very high quality HD materials. All this results come from validation that has been achieved through internationally validated datasets on which subjective tests based on ITU-T BT.500 methodology have been carried out. Pearson correlation coefficient has been calculated to verify the accuracy of PARMENIA and its reliability.

Model and performance of a no-reference quality assessment metric for video streaming

Video streaming via Transmission Control Protocol (TCP) networks has become a popular and highly demanded service, but its quality assessment in both objective and subjective terms has not been properly addressed. In this paper, based on statistical analysis a full analytic model of a no-reference objective metric, namely pause intensity (PI), for video quality assessment is presented. The model characterizes the video playout buffer behavior in connection with the network performance (throughput) and the video playout rate. This allows for instant quality measurement and control without requiring a reference video. PI specifically addresses the need for assessing the quality issue in terms of the continuity in the playout of TCP streaming videos, which cannot be properly measured by other objective metrics such as peak signal-to-noise-ratio, structural similarity, and buffer underrun or pause frequency. The performance of the analytical model is rigidly verified by simulation results and subjective tests using a range of video clips. It is demonstrated that PI is closely correlated with viewers' opinion scores regardless of the vastly different composition of individual elements, such as pause duration and pause frequency which jointly constitute this new quality metric. It is also shown that the correlation performance of PI is consistent and content independent. © 2013 IEEE.

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.