Predicting movie ratings from audience behaviors

We propose a method of representing audience behavior through facial and body motions from a single video stream, and use these features to predict the rating for feature-length movies. This is a very challenging problem as: i) the movie viewing environment is dark and contains views of people at different scales and viewpoints; ii) the duration of feature-length movies is long (80-120 mins) so tracking people uninterrupted for this length of time is still an unsolved problem, and; iii) expressions and motions of audience members are subtle, short and sparse making labeling of activities unreliable. To circumvent these issues, we use an infrared illuminated test-bed to obtain a visually uniform input. We then utilize motion-history features which capture the subtle movements of a person within a pre-defined volume, and then form a group representation of the audience by a histogram of pair-wise correlations over a small-window of time. Using this group representation, we learn our movie rating classifier from crowd-sourced ratings collected by rottentomatoes.com and show our prediction capability on audiences from 30 movies across 250 subjects (> 50 hrs).

An evaluation of crowd counting methods, features and regression models

Existing crowd counting algorithms rely on holistic, local or histogram based features to capture crowd properties. Regression is then employed to estimate the crowd size. Insufficient testing across multiple datasets has made it difficult to compare and contrast different methodologies. This paper presents an evaluation across multiple datasets to compare holistic, local and histogram based methods, and to compare various image features and regression models. A K-fold cross validation protocol is followed to evaluate the performance across five public datasets: UCSD, PETS 2009, Fudan, Mall and Grand Central datasets. Image features are categorised into five types: size, shape, edges, keypoints and textures. The regression models evaluated are: Gaussian process regression (GPR), linear regression, K nearest neighbours (KNN) and neural networks (NN). The results demonstrate that local features outperform equivalent holistic and histogram based features; optimal performance is observed using all image features except for textures; and that GPR outperforms linear, KNN and NN regression

An efficient and robust system for multi-person event detection in real world indoor surveillance scenes

Due to the popularity of security cameras in public places, it is of interest to design an intelligent system that can efficiently detect events automatically. This paper proposes a novel algorithm for multi-person event detection. To ensure greater than real-time performance, features are extracted directly from compressed MPEG video. A novel histogram-based feature descriptor that captures the angles between extracted particle trajectories is proposed, which allows us to capture motion patterns of multi-person events in the video. To alleviate the need for fine-grained annotation, we propose the use of Labelled Latent Dirichlet Allocation, a “weakly supervised” method that allows the use of coarse temporal annotations which are much simpler to obtain. This novel system is able to run at approximately ten times real-time, while preserving state-of-theart detection performance for multi-person events on a 100-hour real-world surveillance dataset (TRECVid SED).

An MRF based abnormal event detection approach using motion and appearance features

Abnormal event detection has attracted a lot of attention in the computer vision research community during recent years due to the increased focus on automated surveillance systems to improve security in public places. Due to the scarcity of training data and the definition of an abnormality being dependent on context, abnormal event detection is generally formulated as a data-driven approach where activities are modeled in an unsupervised fashion during the training phase. In this work, we use a Gaussian mixture model (GMM) to cluster the activities during the training phase, and propose a Gaussian mixture model based Markov random field (GMM-MRF) to estimate the likelihood scores of new videos in the testing phase. Further-more, we propose two new features: optical acceleration, and the histogram of optical flow gradients; to detect the presence of any abnormal objects and speed violations in the scene. We show that our proposed method outperforms other state of the art abnormal event detection algorithms on publicly available UCSD dataset.

Feature selection using expected attainable discrimination

We propose expected attainable discrimination (EAD) as a measure to select discrete valued features for reliable discrimination between two classes of data. EAD is an average of the area under the ROC curves obtained when a simple histogram probability density model is trained and tested on many random partitions of a data set. EAD can be incorporated into various stepwise search methods to determine promising subsets of features, particularly when misclassification costs are difficult or impossible to specify. Experimental application to the problem of risk prediction in pregnancy is described.

Supervised latent dirichlet allocation models for efficient activity representation

Local spatio-temporal features with a Bag-of-visual words model is a popular approach used in human action recognition. Bag-of-features methods suffer from several challenges such as extracting appropriate appearance and motion features from videos, converting extracted features appropriate for classification and designing a suitable classification framework. In this paper we address the problem of efficiently representing the extracted features for classification to improve the overall performance. We introduce two generative supervised topic models, maximum entropy discrimination LDA (MedLDA) and class- specific simplex LDA (css-LDA), to encode the raw features suitable for discriminative SVM based classification. Unsupervised LDA models disconnect topic discovery from the classification task, hence yield poor results compared to the baseline Bag-of-words framework. On the other hand supervised LDA techniques learn the topic structure by considering the class labels and improve the recognition accuracy significantly. MedLDA maximizes likelihood and within class margins using max-margin techniques and yields a sparse highly discriminative topic structure; while in css-LDA separate class specific topics are learned instead of common set of topics across the entire dataset. In our representation first topics are learned and then each video is represented as a topic proportion vector, i.e. it can be comparable to a histogram of topics. Finally SVM classification is done on the learned topic proportion vector. We demonstrate the efficiency of the above two representation techniques through the experiments carried out in two popular datasets. Experimental results demonstrate significantly improved performance compared to the baseline Bag-of-features framework which uses kmeans to construct histogram of words from the feature vectors.

Similarity-based birdcall retrieval from environmental audio

Automated digital recordings are useful for large-scale temporal and spatial environmental monitoring. An important research effort has been the automated classification of calling bird species. In this paper we examine a related task, retrieval of birdcalls from a database of audio recordings, similar to a user supplied query call. Such a retrieval task can sometimes be more useful than an automated classifier. We compare three approaches to similarity-based birdcall retrieval using spectral ridge features and two kinds of gradient features, structure tensor and the histogram of oriented gradients. The retrieval accuracy of our spectral ridge method is 94% compared to 82% for the structure tensor method and 90% for the histogram of gradients method. Additionally, this approach potentially offers a more compact representation and is more computationally efficient.

Generalised features for bird vocalisation retrieval in acoustic recordings

Bioacoustic monitoring has become a significant research topic for species diversity conservation. Due to the development of sensing techniques, acoustic sensors are widely deployed in the field to record animal sounds over a large spatial and temporal scale. With large volumes of collected audio data, it is essential to develop semi-automatic or automatic techniques to analyse the data. This can help ecologists make decisions on how to protect and promote the species diversity. This paper presents generic features to characterize a range of bird species for vocalisation retrieval. In the implementation, audio recordings are first converted to spectrograms using short-time Fourier transform, then a ridge detection method is applied to the spectrogram for detecting points of interest. Based on the detected points, a new region representation are explored for describing various bird vocalisations and a local descriptor including temporal entropy, frequency bin entropy and histogram of counts of four ridge directions is calculated for each sub-region. To speed up the retrieval process, indexing is carried out and the retrieved results are ranked according to similarity scores. The experiment results show that our proposed feature set can achieve 0.71 in term of retrieval success rate which outperforms spectral ridge features alone (0.55) and Mel frequency cepstral coefficients (0.36).

Compact features for birdcall retrieval from environmental acoustic recordings

Bioacoustic data can be used for monitoring animal species diversity. The deployment of acoustic sensors enables acoustic monitoring at large temporal and spatial scales. We describe a content-based birdcall retrieval algorithm for the exploration of large data bases of acoustic recordings. In the algorithm, an event-based searching scheme and compact features are developed. In detail, ridge events are detected from audio files using event detection on spectral ridges. Then event alignment is used to search through audio files to locate candidate instances. A similarity measure is then applied to dimension-reduced spectral ridge feature vectors. The event-based searching method processes a smaller list of instances for faster retrieval. The experimental results demonstrate that our features achieve better success rate than existing methods and the feature dimension is greatly reduced.

Study of carotid arterial plaque stress for symptomatic and asymptomatic patients

Stroke is one of the leading causes of death in the world, resulting mostly from the sudden ruptures of atherosclerosis carotid plaques. Until now, the exact plaque rupture mechanism has not been fully understood, and also the plaque rupture risk stratification. The advanced multi-spectral magnetic resonance imaging (MRI) has allowed the plaque components to be visualized in-vivo and reconstructed by computational modeling. In the study, plaque stress analysis using fully coupled fluid structure interaction was applied to 20 patients (12 symptomatic and 8 asymptomatic) reconstructed from in-vivo MRI, followed by a detailed biomechanics analysis, and morphological feature study. The locally extreme stress conditions can be found in the fibrous cap region, 85% at the plaque shoulder based on the present study cases. Local maximum stress values predicted in the plaque region were found to be significantly higher in symptomatic patients than that in asymptomatic patients (200±43. kPa vs. 127±37. kPa, p=0.001). Plaque stress level, defined by excluding 5% highest stress nodes in the fibrous cap region based on the accumulative histogram of stress experienced on the computational nodes in the fibrous cap, was also significantly higher in symptomatic patients than that in asymptomatic patients (154±32. kPa vs. 111±23. kPa, p<0.05). Although there was no significant difference in lipid core size between the two patient groups, symptomatic group normally had a larger lipid core and a significantly thinner fibrous cap based on the reconstructed plaques using 3D interpolation from stacks of 2D contours. Plaques with a higher stenosis were more likely to have extreme stress conditions upstream of plaque throat. The combined analyses of plaque MR image and plaque stress will advance our understanding of plaque rupture, and provide a useful tool on assessing plaque rupture risk.

On Visual Detection of Highly-occluded Objects for Harvesting Automation in Horticulture

Developing accurate and reliable crop detection algorithms is an important step for harvesting automation in horticulture. This paper presents a novel approach to visual detection of highly-occluded fruits. We use a conditional random field (CRF) on multi-spectral image data (colour and Near-Infrared Reflectance, NIR) to model two classes: crop and background. To describe these two classes, we explore a range of visual-texture features including local binary pattern, histogram of oriented gradients, and learn auto-encoder features. The pro-posed methods are evaluated using hand-labelled images from a dataset captured on a commercial capsicum farm. Experimental results are presented, and performance is evaluated in terms of the Area Under the Curve (AUC) of the precision-recall curves.Our current results achieve a maximum performance of 0.81AUC when combining all of the texture features in conjunction with colour information.

Estimation of fuzzy memberships from histograms

Based on the conclusions drawn in the bijective transformation between possibility and probability, a method is proposed to estimate the fuzzy membership function for pattern recognition purposes. A rational function approximation to the probability density function is obtained from the histogram of a finite (and sometimes very small) number of samples. This function is normalized such that the highest ordinate is one. The parameters representing the rational function are used for classifying the pattern samples based on a max-min decision rule. The method is illustrated with examples.

Computationally Efficient Methods for MDL-Optimal Density Estimation and Data Clustering

The Minimum Description Length (MDL) principle is a general, well-founded theoretical formalization of statistical modeling. The most important notion of MDL is the stochastic complexity, which can be interpreted as the shortest description length of a given sample of data relative to a model class. The exact definition of the stochastic complexity has gone through several evolutionary steps. The latest instantation is based on the so-called Normalized Maximum Likelihood (NML) distribution which has been shown to possess several important theoretical properties. However, the applications of this modern version of the MDL have been quite rare because of computational complexity problems, i.e., for discrete data, the definition of NML involves an exponential sum, and in the case of continuous data, a multi-dimensional integral usually infeasible to evaluate or even approximate accurately. In this doctoral dissertation, we present mathematical techniques for computing NML efficiently for some model families involving discrete data. We also show how these techniques can be used to apply MDL in two practical applications: histogram density estimation and clustering of multi-dimensional data.

Learning Nonlinear Visual Processing from Natural Images

The paradigm of computational vision hypothesizes that any visual function -- such as the recognition of your grandparent -- can be replicated by computational processing of the visual input. What are these computations that the brain performs? What should or could they be? Working on the latter question, this dissertation takes the statistical approach, where the suitable computations are attempted to be learned from the natural visual data itself. In particular, we empirically study the computational processing that emerges from the statistical properties of the visual world and the constraints and objectives specified for the learning process. This thesis consists of an introduction and 7 peer-reviewed publications, where the purpose of the introduction is to illustrate the area of study to a reader who is not familiar with computational vision research. In the scope of the introduction, we will briefly overview the primary challenges to visual processing, as well as recall some of the current opinions on visual processing in the early visual systems of animals. Next, we describe the methodology we have used in our research, and discuss the presented results. We have included some additional remarks, speculations and conclusions to this discussion that were not featured in the original publications. We present the following results in the publications of this thesis. First, we empirically demonstrate that luminance and contrast are strongly dependent in natural images, contradicting previous theories suggesting that luminance and contrast were processed separately in natural systems due to their independence in the visual data. Second, we show that simple cell -like receptive fields of the primary visual cortex can be learned in the nonlinear contrast domain by maximization of independence. Further, we provide first-time reports of the emergence of conjunctive (corner-detecting) and subtractive (opponent orientation) processing due to nonlinear projection pursuit with simple objective functions related to sparseness and response energy optimization. Then, we show that attempting to extract independent components of nonlinear histogram statistics of a biologically plausible representation leads to projection directions that appear to differentiate between visual contexts. Such processing might be applicable for priming, \ie the selection and tuning of later visual processing. We continue by showing that a different kind of thresholded low-frequency priming can be learned and used to make object detection faster with little loss in accuracy. Finally, we show that in a computational object detection setting, nonlinearly gain-controlled visual features of medium complexity can be acquired sequentially as images are encountered and discarded. We present two online algorithms to perform this feature selection, and propose the idea that for artificial systems, some processing mechanisms could be selectable from the environment without optimizing the mechanisms themselves. In summary, this thesis explores learning visual processing on several levels. The learning can be understood as interplay of input data, model structures, learning objectives, and estimation algorithms. The presented work adds to the growing body of evidence showing that statistical methods can be used to acquire intuitively meaningful visual processing mechanisms. The work also presents some predictions and ideas regarding biological visual processing.

Kernel-based spatial-color modeling for fast moving object tracking

Visual tracking has been a challenging problem in computer vision over the decades. The applications of Visual Tracking are far-reaching, ranging from surveillance and monitoring to smart rooms. Mean-shift (MS) tracker, which gained more attention recently, is known for tracking objects in a cluttered environment and its low computational complexity. The major problem encountered in histogram-based MS is its inability to track rapidly moving objects. In order to track fast moving objects, we propose a new robust mean-shift tracker that uses both spatial similarity measure and color histogram-based similarity measure. The inability of MS tracker to handle large displacements is circumvented by the spatial similarity-based tracking module, which lacks robustness to object's appearance change. The performance of the proposed tracker is better than the individual trackers for tracking fast-moving objects with better accuracy.