Interferences in Match Kernels.

We consider the design of an image representation that embeds and aggregates a set of local descriptors into a single vector. Popular representations of this kind include the bag-of-visual-words, the Fisher vector and the VLAD. When two such image representations are compared with the dot-product, the image-to-image similarity can be interpreted as a match kernel. In match kernels, one has to deal with interference, i.e., with the fact that even if two descriptors are unrelated, their matching score may contribute to the overall similarity. We formalise this problem and propose two related solutions, both aimed at equalising the individual contributions of the local descriptors in the final representation. These methods modify the aggregation stage by including a set of per-descriptor weights. They differ by the objective function that is optimised to compute those weights. The first is a "democratisation" strategy that aims at equalising the relative importance of each descriptor in the set comparison metric. The second one involves equalising the match of a single descriptor to the aggregated vector. These concurrent methods give a substantial performance boost over the state of the art in image search with short or mid-size vectors, as demonstrated by our experiments on standard public image retrieval benchmarks.

Label-Embedding for Image Classification.

Attributes act as intermediate representations that enable parameter sharing between classes, a must when training data is scarce. We propose to view attribute-based image classification as a label-embedding problem: each class is embedded in the space of attribute vectors. We introduce a function that measures the compatibility between an image and a label embedding. The parameters of this function are learned on a training set of labeled samples to ensure that, given an image, the correct classes rank higher than the incorrect ones. Results on the Animals With Attributes and Caltech-UCSD-Birds datasets show that the proposed framework outperforms the standard Direct Attribute Prediction baseline in a zero-shot learning scenario. Label embedding enjoys a built-in ability to leverage alternative sources of information instead of or in addition to attributes, such as, e.g., class hierarchies or textual descriptions. Moreover, label embedding encompasses the whole range of learning settings from zero-shot learning to regular learning with a large number of labeled examples.

Good practice in large-scale learning for image classification.

We benchmark several SVM objective functions for large-scale image classification. We consider one-versus-rest, multiclass, ranking, and weighted approximate ranking SVMs. A comparison of online and batch methods for optimizing the objectives shows that online methods perform as well as batch methods in terms of classification accuracy, but with a significant gain in training speed. Using stochastic gradient descent, we can scale the training to millions of images and thousands of classes. Our experimental evaluation shows that ranking-based algorithms do not outperform the one-versus-rest strategy when a large number of training examples are used. Furthermore, the gap in accuracy between the different algorithms shrinks as the dimension of the features increases. We also show that learning through cross-validation the optimal rebalancing of positive and negative examples can result in a significant improvement for the one-versus-rest strategy. Finally, early stopping can be used as an effective regularization strategy when training with online algorithms. Following these "good practices," we were able to improve the state of the art on a large subset of 10K classes and 9M images of ImageNet from 16.7 percent Top-1 accuracy to 19.1 percent.

Asymmetric distances for binary embeddings.

In large-scale query-by-example retrieval, embedding image signatures in a binary space offers two benefits: data compression and search efficiency. While most embedding algorithms binarize both query and database signatures, it has been noted that this is not strictly a requirement. Indeed, asymmetric schemes that binarize the database signatures but not the query still enjoy the same two benefits but may provide superior accuracy. In this work, we propose two general asymmetric distances that are applicable to a wide variety of embedding techniques including locality sensitive hashing (LSH), locality sensitive binary codes (LSBC), spectral hashing (SH), PCA embedding (PCAE), PCAE with random rotations (PCAE-RR), and PCAE with iterative quantization (PCAE-ITQ). We experiment on four public benchmarks containing up to 1M images and show that the proposed asymmetric distances consistently lead to large improvements over the symmetric Hamming distance for all binary embedding techniques.

Iterative quantization: a Procrustean approach to learning binary codes for large-scale image retrieval.

This paper addresses the problem of learning similarity-preserving binary codes for efficient similarity search in large-scale image collections. We formulate this problem in terms of finding a rotation of zero-centered data so as to minimize the quantization error of mapping this data to the vertices of a zero-centered binary hypercube, and propose a simple and efficient alternating minimization algorithm to accomplish this task. This algorithm, dubbed iterative quantization (ITQ), has connections to multiclass spectral clustering and to the orthogonal Procrustes problem, and it can be used both with unsupervised data embeddings such as PCA and supervised embeddings such as canonical correlation analysis (CCA). The resulting binary codes significantly outperform several other state-of-the-art methods. We also show that further performance improvements can result from transforming the data with a nonlinear kernel mapping prior to PCA or CCA. Finally, we demonstrate an application of ITQ to learning binary attributes or "classemes" on the ImageNet data set.

Distance-based image classification: generalizing to new classes at near-zero cost.

We study large-scale image classification methods that can incorporate new classes and training images continuously over time at negligible cost. To this end, we consider two distance-based classifiers, the k-nearest neighbor (k-NN) and nearest class mean (NCM) classifiers, and introduce a new metric learning approach for the latter. We also introduce an extension of the NCM classifier to allow for richer class representations. Experiments on the ImageNet 2010 challenge dataset, which contains over 10(6) training images of 1,000 classes, show that, surprisingly, the NCM classifier compares favorably to the more flexible k-NN classifier. Moreover, the NCM performance is comparable to that of linear SVMs which obtain current state-of-the-art performance. Experimentally, we study the generalization performance to classes that were not used to learn the metrics. Using a metric learned on 1,000 classes, we show results for the ImageNet-10K dataset which contains 10,000 classes, and obtain performance that is competitive with the current state-of-the-art while being orders of magnitude faster. Furthermore, we show how a zero-shot class prior based on the ImageNet hierarchy can improve performance when few training images are available.

A model-based sequence similarity with application to handwritten word spotting.

This paper proposes a novel similarity measure between vector sequences. We work in the framework of model-based approaches, where each sequence is first mapped to a Hidden Markov Model (HMM) and then a measure of similarity is computed between the HMMs. We propose to model sequences with semicontinuous HMMs (SC-HMMs). This is a particular type of HMM whose emission probabilities in each state are mixtures of shared Gaussians. This crucial constraint provides two major benefits. First, the a priori information contained in the common set of Gaussians leads to a more accurate estimate of the HMM parameters. Second, the computation of a similarity between two SC-HMMs can be simplified to a Dynamic Time Warping (DTW) between their mixture weight vectors, which significantly reduces the computational cost. Experiments are carried out on a handwritten word retrieval task in three different datasets-an in-house dataset of real handwritten letters, the George Washington dataset, and the IFN/ENIT dataset of Arabic handwritten words. These experiments show that the proposed similarity outperforms the traditional DTW between the original sequences, and the model-based approach which uses ordinary continuous HMMs. We also show that this increase in accuracy can be traded against a significant reduction of the computational cost.

Aggregating local image descriptors into compact codes.

This paper addresses the problem of large-scale image search. Three constraints have to be taken into account: search accuracy, efficiency, and memory usage. We first present and evaluate different ways of aggregating local image descriptors into a vector and show that the Fisher kernel achieves better performance than the reference bag-of-visual words approach for any given vector dimension. We then jointly optimize dimensionality reduction and indexing in order to obtain a precise vector comparison as well as a compact representation. The evaluation shows that the image representation can be reduced to a few dozen bytes while preserving high accuracy. Searching a 100 million image data set takes about 250 ms on one processor core.

Universal and adapted vocabularies for generic visual categorization.

Generic Visual Categorization (GVC) is the pattern classification problem which consists in assigning labels to an image based on its semantic content. This is a challenging task as one has to deal with inherent object/scene variations as well as changes in viewpoint, lighting and occlusion. Several state-of-the-art GVC systems use a vocabulary of visual terms to characterize images with a histogram of visual word counts. We propose a novel practical approach to GVC based on a universal vocabulary, which describes the content of all the considered classes of images, and class vocabularies obtained through the adaptation of the universal vocabulary using class-specific data. The main novelty is that an image is characterized by a set of histograms - one per class - where each histogram describes whether the image content is best modeled by the universal vocabulary or the corresponding class vocabulary. This framework is applied to two types of local image features: low-level descriptors such as the popular SIFT and high-level histograms of word co-occurrences in a spatial neighborhood. It is shown experimentally on two challenging datasets (an in-house database of 19 categories and the PASCAL VOC 2006 dataset) that the proposed approach exhibits state-of-the-art performance at a modest computational cost.

A probabilistic model of face mapping with local transformations and its application to person recognition.

This paper proposes a new measure of "distance" between faces. This measure involves the estimation of the set of possible transformations between face images of the same person. The global transformation, which is assumed to be too complex for direct modeling, is approximated by a patchwork of local transformations, under a constraint imposing consistency between neighboring local transformations. The proposed system of local transformations and neighboring constraints is embedded within the probabilistic framework of a two-dimensional hidden Markov model. More specifically, we model two types of intraclass variabilities involving variations in facial expressions and illumination, respectively. The performance of the resulting method is assessed on a large data set consisting of four face databases. In particular, it is shown to outperform a leading approach to face recognition, namely, the Bayesian intra/extrapersonal classifier.