Strong Prior Knowledge Research Articles

Integration of prior knowledge into dense image matching for video surveillance

Abstract. Three-dimensional information from dense image matching is a valuable input for a broad range of vision applications. While reliable approaches exist for dedicated stereo setups they do not easily generalize to more challenging camera configurations. In the context of video surveillance the typically large spatial extent of the region of interest and repetitive structures in the scene render the application of dense image matching a challenging task. In this paper we present an approach that derives strong prior knowledge from a planar approximation of the scene. This information is integrated into a graph-cut based image matching framework that treats the assignment of optimal disparity values as a labelling task. Introducing the planar prior heavily reduces ambiguities together with the search space and increases computational efficiency. The results provide a proof of concept of the proposed approach. It allows the reconstruction of dense point clouds in more general surveillance camera setups with wider stereo baselines.

Silhouette estimation.

Silhouettes arise in a variety of imaging scenarios. Pristine silhouettes are often degraded via blurring, detector sampling, and detector noise. We present a maximum a posteriori estimator for the restoration of parameterized facial silhouettes. Extreme dealiasing and dramatic superresolution, well beyond the diffraction limit, are demonstrated through the use of strong prior knowledge.

Fast Methods for Denoising Matrix Completion Formulations, with Applications to Robust Seismic Data Interpolation

Recent SVD-free matrix factorization formulations have enabled rank minimization for systems with millions of rows and columns, paving the way for matrix completion in extremely large-scale applications, such as seismic data interpolation. In this paper, we consider matrix completion formulations designed to hit a target data-fitting error level provided by the user, and we propose an algorithm called LR-BPDN that is able to exploit factorized formulations to solve the corresponding optimization problem. Since practitioners typically have strong prior knowledge about target error level, this innovation makes it easy to apply the algorithm in practice, leaving only the factor rank to be determined. Within the established framework, we propose two extensions that are highly relevant to solving practical challenges of data interpolation. First, we propose a weighted extension that allows known subspace information to improve the results of matrix completion formulations. We show how this weighting can be used in the context of frequency continuation, an essential aspect to seismic data interpolation. Second, we propose matrix completion formulations that are robust to large measurement errors in the available data. We illustrate the advantages of LR-BPDN on collaborative filtering problems using the MovieLens 1M and 10M and Netflix 100M datasets. Then we use the new method, along with its robust and subspace reweighted extensions, to obtain high-quality reconstructions for large-scale seismic interpolation problems with real data, even in the presence of data contamination.

Avoiding Frustrations Of Unprepared Students With Online Quizzes

The hierarchical nature of many degrees enables higher-level courses to build on knowledge that has been developed in earlier courses. However, when students enter with weak prior knowledge, lecturers have to spend time addressing this before starting with the new material. This adds time pressure and frustration to lecturers as well as students who have strong prior knowledge. In this paper, we discuss a strategy that we implemented in order to encourage students to revise or learn prerequisite material at the beginning of a masters level module. Students were asked to take an online quiz on the prerequisite topics. Immediate feedback directed the students to resources which could enhance their knowledge and understanding of the material prior to course commencement. We discuss the multiple benefits this had, for both students and the lecturer, drawing on students written responses to reflective questions about the experience and reflections from the lecturer on the use of online quizzes.

Data-Driven MRSI Spectral Localization Via Low-Rank Component Analysis

Magnetic resonance spectroscopic imaging (MRSI) is a powerful tool capable of providing spatially localized maps of metabolite concentrations. Its utility, however, is often depreciated by spectral leakage artifacts resulting from low spatial resolution measurements through an effort to reduce acquisition times. Though model-based techniques can help circumvent these drawbacks, they require strong prior knowledge, and can introduce additional artifacts when the underlying models are inaccurate. We introduce a novel scheme in which a generative model is estimated from the raw MRSI data via a regularized variational framework that minimizes the model approximation error within a measurement-prescribed subspace. As additional a priori information, our approach relies only upon a measured field inhomogeneity map at high spatial resolution. We demonstrate the feasibility of our approach on both synthetic and experimental data.

Knowing better: The role of prior knowledge and culture in trust in testimony.

This article examined the ability of young children to adapt their trust in testimony in relation to the strength of their prior knowledge across 2 cultures and 2 age groups. Kindergartners and second graders in the United States and Hong Kong (N = 128) viewed pictures of objects and made category judgments about each object: first, in the presence of a teacher who provided labels that conflicted with their prior knowledge, and again, in a pedagogical scenario, when told they were passing the information on to younger learners in the teacher's absence. Results showed that children endorsed more conflicting labels when they had weak (as opposed to strong) prior knowledge about the objects. This effect of knowledge strength was robust across cultures and stronger in the older children. Surprisingly, relative to their Chinese peers, American kindergartners were more willing to endorse conflicting labels when they had strong prior knowledge about the objects. Findings are discussed with respect to the ways in which children are socialized to learn from others across the early primary school years in these 2 very different cultures.

Predicting protein β-sheet contacts using a maximum entropy-based correlated mutation measure

The problem of ab initio protein folding is one of the most difficult in modern computational biology. The prediction of residue contacts within a protein provides a more tractable immediate step. Recently introduced maximum entropy-based correlated mutation measures (CMMs), such as direct information, have been successful in predicting residue contacts. However, most correlated mutation studies focus on proteins that have large good-quality multiple sequence alignments (MSA) because the power of correlated mutation analysis falls as the size of the MSA decreases. However, even with small autogenerated MSAs, maximum entropy-based CMMs contain information. To make use of this information, in this article, we focus not on general residue contacts but contacts between residues in β-sheets. The strong constraints and prior knowledge associated with β-contacts are ideally suited for prediction using a method that incorporates an often noisy CMM. Using contrastive divergence, a statistical machine learning technique, we have calculated a maximum entropy-based CMM. We have integrated this measure with a new probabilistic model for β-contact prediction, which is used to predict both residue- and strand-level contacts. Using our model on a standard non-redundant dataset, we significantly outperform a 2D recurrent neural network architecture, achieving a 5% improvement in true positives at the 5% false-positive rate at the residue level. At the strand level, our approach is competitive with the state-of-the-art single methods achieving precision of 61.0% and recall of 55.4%, while not requiring residue solvent accessibility as an input. http://www2.warwick.ac.uk/fac/sci/systemsbiology/research/software/

Case-Based Instruction in Early Childhood Teacher Preparation: Does It Work?

Case analysis is often used in early childhood teacher education as a constructivist method for developing students' professional skills and knowledge. Although case-based instruction is popular, the professional literature contains little empirical evidence that it effectively helps students develop professional knowledge. Indeed, some empirical evidence suggests case-based instruction may negatively impact learning, especially for students with limited existing knowledge about children's development. This study compared the child development content knowledge gains of two classes of undergraduate preservice early childhood teachers: those taught child development with case-based instruction and those taught the same content with traditional didactic instruction. The study also sought to determine if students' level of prior knowledge interacted with the type of instruction they received. Results indicated that students with strong prior knowledge out-performed classmates regardless of instructional method (i.e., case-based or didactic instruction). Students at all levels of prior knowledge had similar knowledge gains across both case-based and didactic instruction, but data trends suggest students with the least prior knowledge gained the most knowledge with case-based instruction.

Incremental kernel learning for active image retrieval without global dictionaries

In content-based image retrieval context, a classic strategy consists in computing off-line a dictionary of visual features. This visual dictionary is then used to provide a new representation of the data which should ease any task of classification or retrieval. This strategy, based on past research works in text retrieval, is suitable for the context of batch learning, when a large training set can be built either by using a strong prior knowledge of data semantics (like for textual data) or with an expensive off-line pre-computation. Such an approach has major drawbacks in the context of interactive retrieval, where the user iteratively builds the training data set in a semi-supervised approach by providing positive and negative annotations to the system in the relevance feedback loop. The training set is thus built for each retrieval session without any prior knowledge about the concepts of interest for this session. We propose a completely different approach to build the dictionary on-line from features extracted in relevant images. We design the corresponding kernel function, which is learnt during the retrieval session. For each new label, the kernel function is updated with a complexity linear with respect to the size of the database. We propose an efficient active learning strategy for the weakly supervised retrieval method developed in this paper. Moreover this framework allows the combination of features of different types. Experiments are carried out on standard databases, and show that a small dictionary can be dynamically extracted from the features with better performances than a global one.

Extended Phase Field Higher-Order Active Contour Models for Networks

This paper addresses the segmentation from an image of entities that have the form of a `network', i.e. the region in the image corresponding to the entity is composed of branches joining together at junctions, e.g. road or vascular networks. We present new phase field higher-order active contour (HOAC) prior models for network regions, and apply them to the segmentation of road networks from very high resolution satellite images. This is a hard problem for two reasons. First, the images are complex, with much `noise' in the road region due to cars, road markings, etc., while the background is very varied, containing many features that are locally similar to roads. Second, network regions are complex to model, because they may have arbitrary topology. In particular, we address a limitation of a previous model in which network branch width was constrained to be similar to maximum network branch radius of curvature, thereby providing a poor model of networks with straight narrow branches or highly curved, wide branches. We solve this problem by introducing first an additional nonlinear nonlocal HOAC term, and then an additional linear nonlocal HOAC term to improve the computational speed. Both terms allow separate control of branch width and branch curvature, and furnish better prolongation for the same width, but the linear term has several advantages: it is more efficient, and it is able to model multiple widths simultaneously. To cope with the difficulty of parameter selection for these models, we perform a stability analysis of a long bar with a given width, and hence show how to choose the parameters of the energy functions. After adding a likelihood energy, we use both models to extract the road network quasi-automatically from pieces of a QuickBird image, and compare the results to other models in the literature. The state-of-the-art results obtained demonstrate the superiority of our new models, the importance of strong prior knowledge in general, and of the new terms in particular.

Is Relevance Relevant? User Relevance Ratings May Not Predict the Impact of Internet Search on Decision Outcomes

ObjectiveA common measure of Internet search engine effectiveness is its ability to find documents that a user perceives as ‘relevant’. This study sought to test whether user provided relevance ratings for documents retrieved by an Internet search engine correlate with the decision outcome after use of a search engine. Design227 university students were asked to answer four randomly assigned consumer health questions, then to conduct an Internet search on one of two randomly assigned search engines of different performance, and to again answer the question. MeasurementsParticipants were asked to provide a relevance score for each document retrieved as well as a pre and post search answer to each question. ResultsUser relevance rankings had little or no predictive power. Relevance rankings were unable to predict whether the user of a search engine could correctly answer a question after search and could not differentiate between two search engines with statistically different performance in the hands of users. Only when users had strong prior knowledge of the questions, and the decision task was of low complexity, did relevance appear to have modest predictive power. ConclusionsUser provided relevance rankings taken in isolation seem to be of limited to no value when designing a search engine that will be used in a general-purpose setting. Relevance rankings may have a place in situations in which experts provide rankings, and decision tasks are of complexity commensurate with the abilities of the raters. A more natural metric of search engine performance may be a user's ability to accurately complete a task, as this removes the inherent subjectivity of relevance rankings, and provides a direct and repeatable outcome measure which directly correlates with the performance of the search technology in the hands of users.

Linear State-Space Models for Blind Source Separation

We apply a type of generative modelling to the problem of blind source separation in which prior knowledge about the latent source signals, such as time-varying auto-correlation and quasi-periodicity, are incorporated into a linear state-space model. In simulations, we show that in terms of signal-to-error ratio, the sources are inferred more accurately as a result of the inclusion of strong prior knowledge. We explore different schemes of maximum-likelihood optimization for the purpose of learning the model parameters. The Expectation Maximization algorithm, which is often considered the standard optimization method in this context, results in slow convergence when the noise variance is small. In such scenarios, quasi-Newton optimization yields substantial improvements in a range of signal to noise ratios. We analyze the performance of the methods on convolutive mixtures of speech signals.

A Bayesian Framework for Extracting Human Gait Using Strong Prior Knowledge

Extracting full-body motion of walking people from monocular video sequences in complex, real-world environments is an important and difficult problem, going beyond simple tracking, whose satisfactory solution demands an appropriate balance between use of prior knowledge and learning from data. We propose a consistent Bayesian framework for introducing strong prior knowledge into a system for extracting human gait. In this work, the strong prior is built from a simple articulated model having both time-invariant (static) and time-variant (dynamic) parameters. The model is easily modified to cater to situations such as walkers wearing clothing that obscures the limbs. The statistics of the parameters are learned from high-quality (indoor laboratory) data and the Bayesian framework then allows us to "bootstrap" to accurate gait extraction on the noisy images typical of cluttered, outdoor scenes. To achieve automatic fitting, we use a hidden Markov model to detect the phases of images in a walking cycle. We demonstrate our approach on silhouettes extracted from fronto-parallel ("sideways on") sequences of walkers under both high-quality indoor and noisy outdoor conditions. As well as high-quality data with synthetic noise and occlusions added, we also test walkers with rucksacks, skirts, and trench coats. Results are quantified in terms of chamfer distance and average pixel error between automatically extracted body points and corresponding hand-labeled points. No one part of the system is novel in itself, but the overall framework makes it feasible to extract gait from very much poorer quality image sequences than hitherto. This is confirmed by comparing person identification by gait using our method and a well-established baseline recognition algorithm.

Facilitating Children's Comprehension through the Use of Advance Organizers

Recent investigations of Ausubel's advance organizer technique, a prereading instuctional intervention which serves to link new material with existing prior knowledge, have examined the interaction between learner characteristics and the characteristics of an advance organizer. However, this research fails to make eplicit the relationship between the advance organizer and the existing prior knowledge of the learner. The current study investigated the effects on comprehension of verbal and pictorial advance organizers with 8-year-old children having high or low prior knowledge relative to a science topic. In Experiment 1 a verbal advance organizer assisted the comprehension of only the children in the strong prior knowledge group. In Experiment 2 the addition of a pictorial component to the verbal advance organizer facilitated the comprehension of children in the weak prior knowledge group.