In this paper, we propose a semi-supervised learning approach for classifying program (bot) generated web search traffic from that of genuine human users. The work is motivated by the challenge that the enormous amount of search data pose to traditional approaches that rely on fully annotated training samples. We propose a semi-supervised framework that addresses the problem in multiple fronts. First, we use the CAPTCHA technique and simple heuristics to extract from the data logs a large set of training samples with initial labels, though directly using these training data is problematic because the data thus sampled are biased. To tackle this problem, we further develop a semi-supervised learning algorithm to take advantage of the unlabeled data to improve the classification performance. These two proposed algorithms can be seamlessly combined and very cost efficient to scale the training process. In our experiment, the proposed approach showed significant (i.e. 2 : 1) improvement compared to the traditional supervised approach.

With the rapid explosion of video data, compact representation of videos is becoming more and more desirable for efficient browsing and communication, which leads to a number of research works on video summarization in recent years. Among these works, summaries based on a set of still frames are frequently studied and applied due to its high compactness. However, the representativeness of the selected frames, which are taken as the compact representation of the video or video segment, has not been well studied. It is observed that frame representativeness is highly related to the following elements: image quality, user attention measure, visual details, and displaying duration. It is also observed that users have similar tendency in selecting the most representative frame for a certain video segment. In this paper, we developed a method to examine and evaluate the representativeness of video frames based on learning users' perceptive evaluations.

We propose a data-driven approach to estimate the likelihood that an image segment corresponds to a scene object (its "objectness") by comparing it to a large collection of example object regions. We demonstrate that when the application domain is known, for example, in our case activity of daily living (ADL), we can capture the regularity of the domain specific objects using millions of exemplar object regions. Our approach to estimating the objectness of an image region proceeds in two steps: 1) finding the exemplar regions that are the most similar to the input image segment; 2) calculating the objectness of the image segment by combining segment properties, mutual consistency across the nearest exemplar regions, and the prior probability of each exemplar region. In previous work, parametric objectness models were built from a small number of manually annotated objects regions, instead, our data-driven approach uses 5 million object regions along with their metadata information. Results on multiple data sets demonstrates our data-driven approach compared to the existing model based techniques. We also show the application of our approach in improving the performance of object discovery algorithms.

Abstract. Object discovery algorithms group together image regions that originate from the same object. This process is effective when the input collection of images contains a large number of densely sampled views of each object, thereby creating strong connections between nearby views. However, existing approaches are less effective when the input data only provide sparse coverage of object views. We propose an approach for object discovery that addresses this problem. We collect a database of about 5 million product images that capture 1.2 million objects from multiple views. We represent each region in the input image by a "bag" of database object regions. We group input regions together if they share similar "bags of regions." Our approach can correctly discover links between regions of the same object even if they are captured from dramatically different viewpoints. With the help from these added links, our proposed approach can robustly discover object instances even with sparse coverage of the viewpoints.

Abstract

In this paper we propose an image indexing and matching algorithm that relies on selecting distinctive high dimensional features. In contrast with conventional techniques that treated all features equally, we claim that one can benefit significantly from focusing on distinctive features. We propose a bag-of-words algorithm that combines the feature distinctiveness in visual vocabulary generation. Our approach compares favorably with the state of the art in image matching tasks on the University of Kentucky Recognition Benchmark dataset and on an indoor localization dataset. We also show that our approach scales up more gracefully on a large scale Flickr dataset.

In this paper, we propose a data driven approach to firstperson vision. We propose a novel image matching algorithm, named Re-Search, that is designed to cope with selfrepetitive structures and confusing patterns in the indoor environment. This algorithm uses state-of-art image search techniques, and it matches a query image with a two-pass strategy. In the first pass, a conventional image search algorithm is used to search for a small number of images that are most similar to the query image. In the second pass, the retrieval results from the first step are used to discover features that are more distinctive in the local context. We demonstrate and evaluate the Re-Search algorithm in the context of indoor localization, with the illustration of potential applications in object pop-out and data-driven zoom-in.

We propose a new data-driven framework for novel object detection and segmentation, or "object pop-out". Traditionally, this task is approached via background subtraction, which requires continuous observation from a stationary camera. Instead, we consider this an image matching problem. We detect novel objects in the scene using an unordered, sparse database of previously captured images of the same general environment. The problem is formulated in a new image composition framework: 1) given an input image, we find a small set of similar matching images; 2) each of the matches is aligned with the input by proposing a set of homography transformations; 3) regions from different transformed matches are stitched together into a single composite image that best matches the input; 4) the difference between the input and the composite is used to "pop-out" new or changed objects.