Context Models for Understanding Images and Videos

A computer vision system that has to interact in natural language needs to understand the visual appearance of interactions between objects along with the appearance of objects themselves. Relationships between objects are frequently mentioned in queries of tasks like semantic image retrieval, image captioning, visual question answering and natural language object detection. Hence, it is essential to model context between objects for solving these tasks. In the first part of this thesis, we present a technique for detecting an object mentioned in a natural language query. Specifically, we work with referring expressions which are sentences that identify a particular object instance in an image. In many referring expressions, an object is described in relation to another object using prepositions, comparative adjectives, action verbs etc. Our proposed technique can identify both the referred object and the context object mentioned in such expressions. Context is also useful for incrementally understanding scenes and videos. In the second part of this thesis, we propose techniques for searching for objects in an image and events in a video. Our proposed incremental algorithms use the context from previously explored regions to prioritize the regions to explore next. The advantage of incremental understanding is restricting the amount of computation time and/or resources spent for various detection tasks. Our first proposed technique shows how to learn context in indoor scenes in an implicit manner and use it for searching for objects. The second technique shows how explicitly written context rules of one-on-one basketball can be used to sequentially detect events in a game.

Comparative study of human age estimation based on hand-crafted and deep face features

In the past few years, human facial age estimation has drawn a lot of attention in the computer vision and pattern recognition communities because of its important applications in age-based image retrieval, security control and surveillance, biomet- rics, human-computer interaction (HCI) and social robotics. In connection with these investigations, estimating the age of a person from the numerical analysis of his/her face image is a relatively new topic. Also, in problems such as Image Classification the Deep Neural Networks have given the best results in some areas including age estimation. In this work we use three hand-crafted features as well as five deep features that can be obtained from pre-trained deep convolutional neural networks. We do a comparative study of the obtained age estimation results with these features.

Birdcall retrieval from environmental acoustic recordings using image processing

Acoustic recordings of the environment provide an effective means to monitor bird species diversity. To facilitate exploration of acoustic recordings, we describe a content-based birdcall retrieval algorithm. A query birdcall is a region of spectrogram bounded by frequency and time. Retrieval depends on a similarity measure derived from the orientation and distribution of spectral ridges. The spectral ridge detection method caters for a broad range of birdcall structures. In this paper, we extend previous work by incorporating a spectrogram scaling step in order to improve the detection of spectral ridges. Compared to an existing approach based on MFCC features, our feature representation achieves better retrieval performance for multiple bird species in noisy recordings.

Learning Embeddings for Indexing, Retrieval, and Classification, with Applications to Object and Shape Recognition in Image Databases

Nearest neighbor retrieval is the task of identifying, given a database of objects and a query object, the objects in the database that are the most similar to the query. Retrieving nearest neighbors is a necessary component of many practical applications, in fields as diverse as computer vision, pattern recognition, multimedia databases, bioinformatics, and computer networks. At the same time, finding nearest neighbors accurately and efficiently can be challenging, especially when the database contains a large number of objects, and when the underlying distance measure is computationally expensive. This thesis proposes new methods for improving the efficiency and accuracy of nearest neighbor retrieval and classification in spaces with computationally expensive distance measures. The proposed methods are domain-independent, and can be applied in arbitrary spaces, including non-Euclidean and non-metric spaces. In this thesis particular emphasis is given to computer vision applications related to object and shape recognition, where expensive non-Euclidean distance measures are often needed to achieve high accuracy. The first contribution of this thesis is the BoostMap algorithm for embedding arbitrary spaces into a vector space with a computationally efficient distance measure. Using this approach, an approximate set of nearest neighbors can be retrieved efficiently - often orders of magnitude faster than retrieval using the exact distance measure in the original space. The BoostMap algorithm has two key distinguishing features with respect to existing embedding methods. First, embedding construction explicitly maximizes the amount of nearest neighbor information preserved by the embedding. Second, embedding construction is treated as a machine learning problem, in contrast to existing methods that are based on geometric considerations. The second contribution is a method for constructing query-sensitive distance measures for the purposes of nearest neighbor retrieval and classification. In high-dimensional spaces, query-sensitive distance measures allow for automatic selection of the dimensions that are the most informative for each specific query object. It is shown theoretically and experimentally that query-sensitivity increases the modeling power of embeddings, allowing embeddings to capture a larger amount of the nearest neighbor structure of the original space. The third contribution is a method for speeding up nearest neighbor classification by combining multiple embedding-based nearest neighbor classifiers in a cascade. In a cascade, computationally efficient classifiers are used to quickly classify easy cases, and classifiers that are more computationally expensive and also more accurate are only applied to objects that are harder to classify. An interesting property of the proposed cascade method is that, under certain conditions, classification time actually decreases as the size of the database increases, a behavior that is in stark contrast to the behavior of typical nearest neighbor classification systems. The proposed methods are evaluated experimentally in several different applications: hand shape recognition, off-line character recognition, online character recognition, and efficient retrieval of time series. In all datasets, the proposed methods lead to significant improvements in accuracy and efficiency compared to existing state-of-the-art methods. In some datasets, the general-purpose methods introduced in this thesis even outperform domain-specific methods that have been custom-designed for such datasets.

FPGA-based minutia matching for biometric fingerprint image database retrieval

In this paper, a parallel-matching processor architecture with early jump-out (EJO) control is proposed to carry out high-speed biometric fingerprint database retrieval. The processor performs the fingerprint retrieval by using minutia point matching. An EJO method is applied to the proposed architecture to speed up the large database retrieval. The processor is implemented on a Xilinx Virtex-E, and occupies 6,825 slices and runs at up to 65 MHz. The software/hardware co-simulation benchmark with a database of 10,000 fingerprints verifies that the matching speed can achieve the rate of up to 1.22 million fingerprints per second. EJO results in about a 22% gain in computing efficiency.

Task-based image annotation and retrieval

In order to address problems of information overload in digital imagery task domains we have developed an interactive approach to the capture and reuse of image context information. Our framework models different aspects of the relationship between images and domain tasks they support by monitoring the interactive manipulation and annotation of task-relevant imagery. The approach allows us to gauge a measure of a user's intentions as they complete goal-directed image tasks. As users analyze retrieved imagery their interactions are captured and an expert task context is dynamically constructed. This human expertise, proficiency, and knowledge can then be leveraged to support other users in carrying out similar domain tasks. We have applied our techniques to two multimedia retrieval applications for two different image domains, namely the geo-spatial and medical imagery domains. © Springer-Verlag Berlin Heidelberg 2007.

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.

Semantic and context-based retrieval of digital cultural objects [Published in Chinese]

Cultural objects are increasingly generated and stored in digital form, yet effective methods for their indexing and retrieval still remain an important area of research. The main problem arises from the disconnection between the content-based indexing approach used by computer scientists and the description-based approach used by information scientists. There is also a lack of representational schemes that allow the alignment of the semantics and context with keywords and low-level features that can be automatically extracted from the content of these cultural objects. This paper presents an integrated approach to address these problems, taking advantage of both computer science and information science approaches. We firstly discuss the requirements from a number of perspectives: users, content providers, content managers and technical systems. We then present an overview of our system architecture and describe various techniques which underlie the major components of the system. These include: automatic object category detection; user-driven tagging; metadata transform and augmentation, and an expression language for digital cultural objects. In addition, we discuss our experience on testing and evaluating some existing collections, analyse the difficulties encountered and propose ways to address these problems.

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.

Effect of topic domain and task type on Web image searching

Many user studies in Web information searching have found the significant effect of task types on search strategies. However, little attention was given to Web image searching strategies, especially the query reformulation activity despite that this is a crucial part in Web image searching. In this study, we investigated the effects of topic domains and task types on user’s image searching behavior and query reformulation strategies. Some significant differences in user’s tasks specificity and initial concepts were identified among the task domains. Task types are also found to influence participant’s result reviewing behavior and query reformulation strategies.

Modelling users' contextual querying behaviour for web image searching

The rapid growth of visual information on Web has led to immense interest in multimedia information retrieval (MIR). While advancement in MIR systems has achieved some success in specific domains, particularly the content-based approaches, general Web users still struggle to find the images they want. Despite the success in content-based object recognition or concept extraction, the major problem in current Web image searching remains in the querying process. Since most online users only express their needs in semantic terms or objects, systems that utilize visual features (e.g., color or texture) to search images create a semantic gap which hinders general users from fully expressing their needs. In addition, query-by-example (QBE) retrieval imposes extra obstacles for exploratory search because users may not always have the representative image at hand or in mind when starting a search (i.e. the page zero problem). As a result, the majority of current online image search engines (e.g., Google, Yahoo, and Flickr) still primarily use textual queries to search. The problem with query-based retrieval systems is that they only capture users’ information need in terms of formal queries;; the implicit and abstract parts of users’ information needs are inevitably overlooked. Hence, users often struggle to formulate queries that best represent their needs, and some compromises have to be made. Studies of Web search logs suggest that multimedia searches are more difficult than textual Web searches, and Web image searching is the most difficult compared to video or audio searches. Hence, online users need to put in more effort when searching multimedia contents, especially for image searches. Most interactions in Web image searching occur during query reformulation. While log analysis provides intriguing views on how the majority of users search, their search needs or motivations are ultimately neglected. User studies on image searching have attempted to understand users’ search contexts in terms of users’ background (e.g., knowledge, profession, motivation for search and task types) and the search outcomes (e.g., use of retrieved images, search performance). However, these studies typically focused on particular domains with a selective group of professional users. General users’ Web image searching contexts and behaviors are little understood although they represent the majority of online image searching activities nowadays. We argue that only by understanding Web image users’ contexts can the current Web search engines further improve their usefulness and provide more efficient searches. In order to understand users’ search contexts, a user study was conducted based on university students’ Web image searching in News, Travel, and commercial Product domains. The three search domains were deliberately chosen to reflect image users’ interests in people, time, event, location, and objects. We investigated participants’ Web image searching behavior, with the focus on query reformulation and search strategies. Participants’ search contexts such as their search background, motivation for search, and search outcomes were gathered by questionnaires. The searching activity was recorded with participants’ think aloud data for analyzing significant search patterns. The relationships between participants’ search contexts and corresponding search strategies were discovered by Grounded Theory approach. Our key findings include the following aspects: - Effects of users' interactive intents on query reformulation patterns and search strategies - Effects of task domain on task specificity and task difficulty, as well as on some specific searching behaviors - Effects of searching experience on result expansion strategies A contextual image searching model was constructed based on these findings. The model helped us understand Web image searching from user perspective, and introduced a context-aware searching paradigm for current retrieval systems. A query recommendation tool was also developed to demonstrate how users’ query reformulation contexts can potentially contribute to more efficient searching.