Yandong Guo

MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition

In this paper, we design a benchmark task and provide the associated datasets for recognizing face images and link them to corresponding entity keys in a knowledge base. More specifically, we propose a benchmark task to recognize one million celebrities from their face images, by using all the possibly collected face images of this individual on the web as training data. The rich information provided by the knowledge base helps to conduct disambiguation and improve the recognition accuracy, and contributes to various real-world applications, such as image captioning and news video analysis. Associated with this task, we design and provide concrete measurement set, evaluation protocol, as well as training data. We also present in details our experiment setup and report promising baseline results. Our benchmark task could lead to one of the largest classification problems in computer vision. To the best of our knowledge, our training dataset, which contains 10M images in version 1, is the largest publicly available one in the world.

High dimensional regression using the sparse matrix transform (SMT)

Regression from high dimensional observation vectors is particularly difficult when training data is limited. More specifically, if the number of sample vectors n is less than dimension of the sample vectors p, then accurate regression is difficult to perform without prior knowledge of the data covariance. In this paper, we propose a novel approach to high dimensional regression for application when n ≪ p. The approach works by first decorrelating the high dimensional observation vector using the sparse matrix transform (SMT) estimate of the data covariance. Then the decorrelated observations are used in a regularized regression procedure such as Lasso or shrinkage. Numerical results demonstrate that the proposed regression approach can significantly improve the prediction accuracy, especially when n is small and the signal to be predicted lies in the subspace of the observations corresponding to the small eigenvalues.

Binary image compression using conditional entropy-based dictionary design and indexing

The JBIG2 standard is widely used for binary document image compression primarily because it achieves much higher compression ratios than conventional facsimile encoding standards, such as T.4, T.6, and T.82 (JBIG1). A typical JBIG2 encoder works by first separating the document into connected components, or symbols. Next it creates a dictionary by encoding a subset of symbols from the image, and finally it encodes all the remaining symbols using the dictionary entries as a reference. In this paper, we propose a novel method for measuring the distance between symbols based on a conditionalentropy estimation (CEE) distance measure. The CEE distance measure is used to both index entries of the dictionary and construct the dictionary. The advantage of the CEE distance measure, as compared to conventional measures of symbol similarity, is that the CEE provides a much more accurate estimate of the number of bits required to encode a symbol. In experiments on a variety of documents, we demonstrate that the incorporation of the CEE distance measure results in approximately a 14% reduction in the overall bitrate of the JBIG2 encoded bitstream as compared to the best conventional dissimilarity measures.

Dynamic hierarchical dictionary design for multi-page binary document image compression

The JBIG2 standard is widely used for binary document image compression primarily because it achieves much higher compression ratios than conventional facsimile encoding standards. In this paper, we propose a dynamic hierarchical dictionary design method (DH) for multi-page binary document image compression with JBIG2. Our DH method outperforms other methods for multi-page compression by utilizing the information redundancy among pages with the following technologies. First, we build a hierarchical dictionary to keep more information per page for future usage. Second, we dynamically update the dictionary in memory to keep as much information as possible subject to the memory constraint. Third, we incorporate our conditional entropy estimation algorithm to utilize the saved information more effectively. Our experimental results show that the compression ratio improvement by our DH method is about 15% compared to the best existing multi-page encoding method.

Adaptive video presentation for small display while maximize visual information

In this paper we focus our attention on solving the contradiction that it is more and more popular to watch videos through mobile devices and there is an explosive growth of mobile devices with multimedia applications but the display sizes of mobile devices are limited and heterogeneous. We present an intact and generic framework to adapt video presentation (AVP). A novel method for choosing the optimal cropped region is introduced to minimize the information loss over adapting video presentation. In order to ameliorate the output stream, we make use of a group of filters for tracking, smoothing and virtual camera controlling. Experiments indicate that our approach is able to achieve satisfactory results and has obvious superiority especially when the display size is pretty small.

Image quality evaluation using image quality ruler and graphical model

Quantifying image quality through subjective evaluation is very critical to image quality evaluation. Using the image quality ruler method, an average score per stimulus can be easily obtained in the unit of Just Noticeable Differences (JNDs). However, it requires a large number of subjects, since pure averaging does not consider the different judging quality of different subjects. In this paper, we propose an image quality evaluation framework using the image quality ruler method with a statistical model. By incorporating this model, we consider the quality score, the expertise of the subjects, and the difficulty of image rating task as three hidden variables. Then we use expectation-maximization (EM) to estimate these hidden variables. From our experimental results, we show that our method provides reliable results without using a large number of subjects. Preliminary results also demonstrate that the estimates of the parameters can guide us to better distribute the valuable human resources used to conduct psychophysical experiments.

Text line detection based on cost optimized local text line direction estimation

Text line detection is a critical step for applications in document image processing. In this paper, we propose a novel text line detection method. First, the connected components are extracted from the image as symbols. Then, we estimate the direction of the text line in multiple local regions. This estimation is, for the first time, to our knowledge, formulated in a cost optimization framework. We also propose an efficient way to solve this optimization problem. Afterwards, we consider symbols as nodes in a graph, and connect symbols based on the local text line direction estimation results. Last, we detect the text lines by separating the graph into subgraphs according to the nodes’ connectivities. Preliminary experimental results demonstrate that our proposed method is very robust to non-uniform skew within text lines, variability of font sizes, and complex structures of layout. Our new method works well for documents captured with flat-bed and sheet-fed scanners, mobile phone cameras, and with other general imaging assets.

Model-Based Iterative Restoration for Binary Document Image Compression with Dictionary Learning

The inherent noise in the observed (e.g., scanned) binary document image degrades the image quality and harms the compression ratio through breaking the pattern repentance and adding entropy to the document images. In this paper, we design a cost function in Bayesian framework with dictionary learning. Minimizing our cost function produces a restored image which has better quality than that of the observed noisy image, and a dictionary for representing and encoding the image. After the restoration, we use this dictionary (from the same cost function) to encode the restored image following the symbol-dictionary framework by JBIG2 standard with the lossless mode. Experimental results with a variety of document images demonstrate that our method improves the image quality compared with the observed image, and simultaneously improves the compression ratio. For the test images with synthetic noise, our method reduces the number of flipped pixels by 48.2% and improves the compression ratio by 36.36% as compared with the best encoding methods. For the test images with real noise, our method visually improves the image quality, and outperforms the cutting-edge method by 28.27% in terms of the compression ratio.

Online image classification under monotonic decision boundary constraint

Image classification is a prerequisite for copy quality enhancement in all-in-one (AIO) device that comprises a printer and scanner, and which can be used to scan, copy and print. Different processing pipelines are provided in an AIO printer. Each of the processing pipelines is designed specifically for one type of input image to achieve the optimal output image quality. A typical approach to this problem is to apply Support Vector Machine to classify the input image and feed it to its corresponding processing pipeline. The online training SVM can help users to improve the performance of classification as input images accumulate. At the same time, we want to make quick decision on the input image to speed up the classification which means sometimes the AIO device does not need to scan the entire image to make a final decision. These two constraints, online SVM and quick decision, raise questions regarding: 1) what features are suitable for classification; 2) how we should control the decision boundary in online SVM training. This paper will discuss the compatibility of online SVM and quick decision capability.