Decomposition-Based Transfer Distance Metric Learning for Image Classification

Abstract

Distance metric learning (DML) is a critical factor for image analysis and pattern recognition. To learn a robust distance metric for a target task, we need abundant side information (i.e., the similarity/dissimilarity pairwise constraints over the labeled data), which is usually unavailable in practice due to the high labeling cost. This paper considers the transfer learning setting by exploiting the large quantity of side information from certain related, but different source tasks to help with target metric learning (with only a little side information). The state-of-the-art metric learning algorithms usually fail in this setting because the data distributions of the source task and target task are often quite different. We address this problem by assuming that the target distance metric lies in the space spanned by the eigenvectors of the source metrics (or other randomly generated bases). The target metric is represented as a combination of the base metrics, which are computed using the decomposed components of the source metrics (or simply a set of random bases); we call the proposed method, decomposition-based transfer DML (DTDML). In particular, DTDML learns a sparse combination of the base metrics to construct the target metric by forcing the target metric to be close to an integration of the source metrics. The main advantage of the proposed method compared with existing transfer metric learning approaches is that we directly learn the base metric coefficients instead of the target metric. To this end, far fewer variables need to be learned. We therefore obtain more reliable solutions given the limited side information and the optimization tends to be faster. Experiments on the popular handwritten image (digit, letter) classification and challenge natural image annotation tasks demonstrate the effectiveness of the proposed method.

Image Dehazing Using Guided Image Filter

“…Methods presented in earlier studies had required multiple images to perform dehazing. For example, polarization-based methods [3,4,5] use the polarization property of scattered light to restore the scene depth information from two or more images taken with different degrees of polarization. Similarly, in [6,7], multiple images of the same scene are captured under different weather conditions to be used as reference.…”

A new transferred feature selection algorithm for customer identification

“…It has been proposed to deal with the situations where data are difficult to collect in the target domain while auxiliary data in the source domain are readily available or relatively easy to collect. Previous research has shown that transfer learning can improve modeling accuracy in many areas such as text mining [11], image analysis [27]. To the best of our knowledge, no reported research has applied transfer learning to solve the class imbalance issue in feature selection.…”

Manifold Regularized Transfer Distance Metric Learning

“…It is necessary to exploit abundant of side information like pairwise constraints to learn a robust and reliable distance metric [2,3]. Let D = {(x l i , x l j , y i j )} n l i, j=1 denotes the labeled training set for the target task, wherein x i , x j ∈ R d and y i j = ±1 indicates x l i and x l i are similar/dissimilar to each other.…”

“…However, when the number of labeled data n l is small, such a simple regularization is often insufficient to control the model complexity. The recently proposed decomposition based TDM-L (DTDML) [2] algorithm is superior to the previous TMDL approaches in that much fewer variables are needed to be learned. Given the m source tasks, we assume there are large amount of n u unlabeled data {x u i , x u j }, as well as m different but related source tasks with abundant labeled training data…”

“…By introducing the regularizer (3) in (1), and adopting the decomposition based metric learning strategy in [2], we obtain the following optimization problem for our MTDML: arg min…”