Leveraging Multiple Features for Image Retrieval and Matching

The goal of image retrieval and matching is to find and locate object instances in images from a large-scale image database. While visual features are abundant, how to combine them to improve performance by individual features remains a challenging task. In this work, we focus on leveraging multiple features for accurate and efficient image retrieval and matching. We first propose two graph-based approaches to rerank initially retrieved images for generic image retrieval. In the graph, vertices are images while edges are similarities between image pairs. Our first approach employs a mixture Markov model based on a random walk model on multiple graphs to fuse graphs. We introduce a probabilistic model to compute the importance of each feature for graph fusion under a naive Bayesian formulation, which requires statistics of similarities from a manually labeled dataset containing irrelevant images. To reduce human labeling, we further propose a fully unsupervised reranking algorithm based on a submodular objective function that can be efficiently optimized by greedy algorithm. By maximizing an information gain term over the graph, our submodular function favors a subset of database images that are similar to query images and resemble each other. The function also exploits the rank relationships of images from multiple ranked lists obtained by different features. We then study a more well-defined application, person re-identification, where the database contains labeled images of human bodies captured by multiple cameras. Re-identifications from multiple cameras are regarded as related tasks to exploit shared information. We apply a novel multi-task learning algorithm using both low level features and attributes. A low rank attribute embedding is joint learned within the multi-task learning formulation to embed original binary attributes to a continuous attribute space, where incorrect and incomplete attributes are rectified and recovered. To locate objects in images, we design an object detector based on object proposals and deep convolutional neural networks (CNN) in view of the emergence of deep networks. We improve a Fast RCNN framework and investigate two new strategies to detect objects accurately and efficiently: scale-dependent pooling (SDP) and cascaded rejection classifiers (CRC). The SDP improves detection accuracy by exploiting appropriate convolutional features depending on the scale of input object proposals. The CRC effectively utilizes convolutional features and greatly eliminates negative proposals in a cascaded manner, while maintaining a high recall for true objects. The two strategies together improve the detection accuracy and reduce the computational cost.

The acquisition of adjunct control: grammar and processing

This dissertation uses children’s acquisition of adjunct control as a case study to investigate grammatical and performance accounts of language acquisition. In previous research, children have consistently exhibited non-adultlike behavior for sentences with adjunct control. To explain children’s behavior, several different grammatical accounts have been proposed, but evidence for these accounts has been inconclusive. In this dissertation, I take two approaches to account for children’s errors. First, I spell out the predictions of previous grammatical accounts, and test these predictions after accounting for some methodological concerns that might have influenced children’s behavior in previous studies. While I reproduce the non-adultlike behavior observed in previous studies, the predictions of previous grammatical accounts are not borne out, suggesting that extragrammatical factors are needed to explain children’s behavior. Next, I consider the role of two different types of extragrammatical factors in predicting children’s non-adultlike behavior. With a new task designed to address the task demands in previous studies, children exhibit significantly higher accuracy than with previous tasks. This suggests that children’s behavior has been influenced by task- specific processing factors. In addition to the task, I also test the predictions of a similarity-based interference account, which links children’s errors to the same memory mechanisms involved in sentence processing difficulties observed in adults. These predictions are borne out, supporting a more continuous developmental trajectory as children’s processing mechanisms become more resistant to interference. Finally, I consider how children’s errors might influence their acquisition of adjunct control, given the distribution in the linguistic input. I discuss the results of a corpus analysis, including the possibility that adjunct control could be learned from the input. The kinds of information that could be useful to a learner become much more limited, however, after considering the processing limitations that would interfere with the representations available to the learner.

Facial and expression recognition for the blind using computer vision

Gemstone Team FACE