The aim of this paper is to address recognition of natural human actions in diverse and realistic video settings. This challenging but important subject has mostly been ignored in the past due to several problems one of which is the lack of realistic and annotated video datasets. Our first contribution is to address this limitation and to investigate the use of movie scripts for automatic annotation of human actions in videos. We evaluate alternative methods for action retrieval from scripts and show benefits of a text-based classifier. Using the retrieved action samples for visual learning, we next turn to the problem of action classification in video. We present a new method for video classification that builds upon and extends several recent ideas including local space-time features, space-time pyramids and multichannel non-linear SVMs. The method is shown to improve state-of-the-art results on the standard KTH action dataset by achieving 91.8 % accuracy. Given the inherent problem of noisy labels in automatic annotation, we particularly investigate and show high tolerance of our method to annotation errors in the training set. We finally apply the method to learning and classifying challenging action classes in movies and show promising results. 1.

Image category recognition is important to access visual information on the level of objects and scene types. So far, intensity-based descriptors have been widely used for feature extraction at salient points. To increase illumination invariance and discriminative power, color descriptors have been proposed. Because many different descriptors exist, a structured overview is required of color invariant descriptors in the context of image category recognition. Therefore, this paper studies the invariance properties and the distinctiveness of color descriptors (software to compute the color descriptors from this paper is available from

State-of-the-art image classification methods require an intensive learning/training stage (using SVM, Boosting, etc.) In contrast, non-parametric Nearest-Neighbor (NN) based image classifiers require no training time and have other favorable properties. However, the large performance gap between these two families of approaches rendered NNbased image classifiers useless. We claim that the effectiveness of non-parametric NNbased image classification has been considerably undervalued. We argue that two practices commonly used in image classification methods, have led to the inferior performance of NN-based image classifiers: (i) Quantization of local image descriptors (used to generate “bags-of-words”, codebooks). (ii) Computation of ‘Image-to-Image ’ distance, instead of ‘Image-to-Class ’ distance. We propose a trivial NN-based classifier – NBNN, (Naive-Bayes Nearest-Neighbor), which employs NNdistances in the space of the local image descriptors (and not in the space of images). NBNN computes direct ‘Imageto-Class’ distances without descriptor quantization. We further show that under the Naive-Bayes assumption, the theoretically optimal image classifier can be accurately approximated by NBNN. Although NBNN is extremely simple, efficient, and requires no learning/training phase, its performance ranks among the top leading learning-based image classifiers. Empirical comparisons are shown on several challenging databases (Caltech-101,Caltech-256 and Graz-01). 1.

Abstract. This paper concerns recognition of human actions under view changes. We explore self-similarities of action sequences over time and observe the striking stability of such measures across views. Building upon this key observation we develop an action descriptor that captures the structure of temporal similarities and dissimilarities within an action sequence. Despite this descriptor not being strictly view-invariant, we provide intuition and experimental validation demonstrating the high stability of self-similarities under view changes. Self-similarity descriptors are also shown stable under action variations within a class as well as discriminative for action recognition. Interestingly, self-similarities computed from different image features possess similar properties and can be used in a complementary fashion. Our method is simple and requires neither structure recovery nor multi-view correspondence estimation. Instead, it relies on weak geometric properties and combines them with machine learning for efficient cross-view action recognition. The method is validated on three public datasets, it has similar or superior performance compared to related methods and it performs well even in extreme conditions such as when recognizing actions from top views while using side views for training only. 1

This paper studies automatic image classification by modeling soft-assignment in the popular codebook model. The codebook model describes an image as a bag of discrete visual words selected from a vocabulary, where the frequency distributions of visual words in an image allow classification. One inherent component of the codebook model is the assignment of discrete visual words to continuous image features. Despite the clear mismatch of this hard assignment with the nature of continuous features, the approach has been applied successfully for some years. In this paper we investigate four types of soft-assignment of visual words to image features. We demonstrate that explicitly modeling visual word assignment ambiguity improves classification performance compared to the hard-assignment of the traditional codebook model. The traditional codebook model is compared against our method for five well-known datasets: 15 natural scenes, Caltech-101, Caltech-256, and Pascal VOC 2007/2008. We demonstrate that large codebook vocabulary sizes completely deteriorate the performance of the traditional model, whereas the proposed model performs consistently. Moreover, we show that our method profits in high-dimensional feature spaces and reaps higher benefits when increasing the number of image categories.

Concept classification is important to access visual information on the level of objects and scene types. So far, intensity-based features have been widely used. To increase discriminative power, color features have been proposed only recently. As many features exist, a structured overview is required of color features in the context of concept classification. Therefore, this paper studies 1. the invariance properties and 2. the distinctiveness of color features in a structured way. The invariance properties of color features with respect to photometric changes are summarized. The distinctiveness of color features is assessed experimentally using an image and a video benchmark: the PASCAL VOC Challenge 2007 and the Mediamill Challenge. Because color features cannot be studied independently from the points at which they are extracted, different point sampling strategies based on Harris-Laplace salient points, dense sampling and the spatial pyramid are also studied. From the experimental results, it can be derived that invariance to light intensity changes and light color changes affects concept classification. The results reveal further that the usefulness of invariance is concept-specific.

This paper addresses recognition of human actions under view changes. We explore self-similarities of action sequences over time and observe the striking stability of such measures across views. Building upon this key observation, we develop an action descriptor that captures the structure of temporal similarities and dissimilarities within an action sequence. Despite this temporal self-similarity descriptor not being strictly view-invariant, we provide intuition and experimental validation demonstrating its high stability under view changes. Self-similarity descriptors are also shown stable under performance variations within a class of actions, when individual speed fluctuations are ignored. If required, such fluctuations between two different instances of the same action class can be explicitly recovered with dynamic time warping, as will be demonstrated, to achieve cross-view action synchronization. More central to present work, temporal ordering of local selfsimilarity descriptors can simply be ignored within a bag-offeatures type of approach. Sufficient action discrimination is still retained this way to build a view-independent action recognition system. Interestingly, self-similarities computed from different image features possess similar properties and can be used in a complementary fashion. Our method is simple and requires neither structure recovery nor multi-view correspondence estimation. Instead, it relies on weak geometric properties and combines them with machine learning for efficient cross-view action recognition. The method is validated on three public datasets. It has similar or superior performance compared to related methods and it performs well even in extreme conditions such as when recognizing actions from top views while using side views only for training.

In this paper we describe our TRECVID 2008 video retrieval experiments. The MediaMill team participated in three tasks: concept detection, automatic search, and interactive search. Rather than continuing to increase the number of concept detectors available for retrieval, our TRECVID 2008 experiments focus on increasing the robustness of a small set of detectors. To that end, our concept detection experiments emphasize in particular the role of sampling, the value of color invariant features, the influence of codebook construction, and the effectiveness of kernel-based learning parameters. For retrieval, a robust but limited set of concept detectors necessitates the need to rely on as many auxiliary information channels as possible. Therefore, our automatic search experiments focus on predicting which information channel to trust given a certain topic, leading to a novel framework for predictive video retrieval. To improve the video retrieval results further, our interactive search experiments investigate the roles of visualizing preview results for a certain browse-dimension and active learning mechanisms that learn to solve complex search topics by analysis from user browsing behavior. The 2008 edition of the TRECVID benchmark has been the most successful MediaMill participation to date, resulting in the top ranking for both concept detection and interactive search, and a runner-up ranking for automatic retrieval. Again a lot has been learned during this year’s TRECVID campaign; we highlight the most important lessons at the end of this paper. 1

In this paper we describe our TRECVID 2009 video retrieval experiments. The MediaMill team participated in three tasks: concept detection, automatic search, and interactive search. Starting point for the MediaMill concept detection approach is our top-performing bag-of-words system of last year, which uses multiple color descriptors, codebooks with soft-assignment, and kernel-based supervised learning. We improve upon this baseline system by exploring two novel research directions. Firstly, we study a multi-modal extension by the inclusion of 20 audio concepts and fusing using two novel multi-kernel supervised learning methods. Secondly, with the help of recently proposed algorithmic refinements of bag-of-words, a bag-of-words GPU implementation, and compute clusters, we scale-up the amount of visual information analyzed by an order of magnitude, to a total of 1,000,000 i-frames. Our experiments evaluate the merit of these new components, ultimately leading to 64 robust concept detectors for video retrieval. For retrieval, a robust but limited set of concept detectors necessitates the need to rely on as many auxiliary information channels as possible. For automatic search we therefore explore how we can learn to rank various information channels simultaneously to maximize video search results for a given topic. To improve the video retrieval results further, our interactive search experiments investigate the roles of visualizing preview results for a certain browse-dimension and relevance feedback mechanisms that learn to solve complex search topics by analysis from user browsing behavior. The 2009 edition of the TRECVID benchmark has again been a fruitful participation for the MediaMill team, resulting in the top ranking for both concept detection and interactive search. 1

Abstract not found