Di Wen

Robust Image Restoration via Adaptive Low-Rank Approximation and Joint Kernel Regression

In recent years, image priors based on nonlocal self-similarity and low-rank approximation have been proven as powerful tools for image restoration. Many restoration methods group similar patches as a matrix and recover the underlying low-rank structure from the corrupted matrix via rank minimization. However, both the nonlocally redundant and low-rank properties are highly content dependent, and whether they can faithfully characterize a wide range of natural images still remains unclear. In this paper, we analyze these two properties and provide quantifications of them in a data-driven and parametric way, respectively, obtaining the new measures of regional redundancy and nonlocal patch rank. Leveraging these prior leads to an adaptive image restoration method with content-awareness. In particular, our method iteratively removes outliers and recovers latent fine details. To handle outliers, we propose an adaptive low-rank and sparse matrix approximation algorithm to encourage the estimated nonlocal rank in the patch matrix. The guidance of regional redundancy further gives rise to the “denoise” quality. In the detail recovery step, we propose an adaptive joint kernel regression algorithm using the redundancy measure to determine the confidence of each regression group. It also bridges the gap between our online and offline dictionary learning schemes. Experiments on synthetic and real-world images show the efficacy of our method in image deblurring and super-resolution tasks, especially when subject to practical outliers such as rain drops.

An effective video text tracking algorithm based on sift feature and geometric constraint

Video text provides important clues for semantic-based video analysis, indexing and retrieval. And text tracking is performed to locate specific text information across video frames and enhance text segmentation and recognition over time. This paper presents a multilingual video text tracking algorithm based on the extraction and tracking of Scale Invariant Feature Transform (SIFT) features description through video frames. SIFT features are extracted from video frames to correspond the region of interests across frames. Meanwhile, a global matching method using geometric constraint is proposed to decrease false matches, which effectively improves the accuracy and stability of text tracking results. Based on the correct matches, the motion of text is estimated in adjacent frames and a match score of text is calculated to determine Text Change Boundary (TCB). Experimental results on a large number of video frames show that the proposed text tracking algorithm is robust to different text forms, including multilingual captions, credits, scene texts with shift, rotation and scale change, under complex backgrounds and light changing.

Document digitization technology and its application for digital library in China

We introduce the research of document digitization technology and its applications for constructing digital libraries in China. We focus on two major objectives of document digitization technologies: performance and efficiency. Taking the most representative TH-OCR product as an example, the up-to-date research achievements on both kernel OCR technologies and peripheral technologies in China are presented. The kernel technologies include high performance multilingual (Chinese, Japanese, Korean and English) text recognition, layout analysis, understanding and reconstruction; the peripheral technologies include the network document digitization workflow and intelligent proofreading, which greatly improve the efficiency. The applications of TH-OCR has two types of final output digital documents, one is the reconstructed electronic document with full text and layout information of the original paper-based document, the other is the multilevel document with OCR output text layer under the image layer. Numerous applications indicate that current technologies can greatly facilitate the mass-volume digitization labour in building digital library infrastructure.

A new biologically inspired active appearance model for face age estimation by using local ordinal ranking

In this paper, a new facial feature called Biologically Inspired Active Appearance Model (BIAAM) is proposed for face age estimation by using a novel age function learning algorithm, called Local Ordinal Ranking (LOR). In BIAAM, appearance variations are encoded by extracting Bio Inspired Feature from normalized shape-free images with a mean shape mask. The proposed LOR divides the training set into several groups according to age labels and applies Ordinal Hyperplanes Ranker for each group to determine the final predicting age. A multiple linear regression function is used to decide which group a query sample belongs to. Experimental evaluation on the FG-NET aging database with mean absolute error 4.18 years demonstrates that our method outperforms other state-of-the-art algorithms.

A general framework for multicharacter segmentation and its application in recognizing multilingual Asian documents

In this paper we propose a general framework for character segmentation in complex multilingual documents, which is an endeavor to combine the traditionally separated segmentation and recognition processes into a cooperative system. The framework contains three basic steps: Dissection, Local Optimization and Global Optimization, which are designed to fuse various properties of the segmentation hypotheses hierarchically into a composite evaluation to decide the final recognition results. Experimental results show that this framework is general enough to be applied in variety of documents. A sample system based on this framework to recognize Chinese, Japanese and Korean documents and experimental performance is reported finally.

Person-based video summarization and retrieval by tracking and clustering temporal face sequences

People are often the most important subjects in videos. It is highly desired to automatically summarize the occurrences of different people in a large collection of video and quickly find the video clips containing a particular person among them. In this paper, we present a person-based video summarization and retrieval system named VideoWho which extracts temporal face sequences in videos and groups them into clusters, with each cluster containing video clips of the same person. This is accomplished based on advanced face detection and tracking algorithms, together with a semisupervised face clustering approach. The system achieved good clustering accuracy when tested on a hybrid video set including home video, TV plays and movies. On top of this technology, a number of applications can be built, such as automatic summarization of major characters in videos, person-related video search on the Internet and personalized UI systems etc.

Face age estimation by using Bisection Search Tree

Age estimation via face images has recently attracted a lot of researches in computer vision, due to its many potential applications. In this paper, a novel Bisection Search Tree (BST) algorithm is proposed for face age estimation, based on the idea of Divide and Conquer. Different from those conventional classification or regression approaches, the BST first constructs a binary tree according to the whole age range of training set, and then learns decision functions for all non-leaf nodes to determine which child node a test sample will be passed to. Gabor wavelet face representation and dimensionality reduction by using Linear Discriminative Analysis are also adopted in this paper. Experimental results on two public aging databases, MORPH-II and MEDS-II, show that the BST method is effective for age estimation and outperforms other state-of-the-art approaches.

Generating face images under multiple illuminations based on a single front-lighted sample without 3D models

In this paper, we propose a novel generative method which could generate images with different illuminations by using a single front-lighted sample. The generative method is based on the linear Lambertian property and requires a bootstrap set with multiple subjects and specific illuminations for each subject. During the generation process, we also propose a scale decomposition method to retain the identity details between the generated image and the original sample. Unlike most of the generative methods, the proposed method does NOT require the 3D model. On the other hand, the generative method is proved flexible to use because it could be implemented altogether with the existing preprocessing methods. Experiment on extended Yale B and FRGC 2.0 databases shows that the generated images could diversify the illuminations of the gallery set, thus improving the recognition performance1.

Visual similarity based document layout analysis

In this paper, a visual similarity based document layout analysis (DLA) scheme is proposed, which by using clustering strategy can adaptively deal with documents in different languages, with different layout structures and skew angles. Aiming at a robust and adaptive DLA approach, the authors first manage to find a set of representative filters and statistics to characterize typical texture patterns in document images, which is through a visual similarity testing process. Texture features are then extracted from these filters and passed into a dynamic clustering procedure, which is called visual similarity clustering. Finally, text contents are located from the clustered results. Benefit from this scheme, the algorithm demonstrates strong robustness and adaptability in a wide variety of documents, which previous traditional DLA approaches do not possess.

Development of Recognition Engine for Baby Faces

Existing face recognition approaches are mostly developed based on adult faces which may not work well in distinguishing faces of kids. Especially, baby faces tend to have common features such as round cheeks and chins, so that current face recognition engines often fail to differentiate them. In this paper, we present methods for discriminating baby faces from adult faces, and for training a special engine to recognize faces of different babies. To achieve these, we collected a huge number of baby face images and developed a software system to annotate the image database. Experimental results prove that the trained baby face recognizer achieves dramatic improvement on differentiating baby faces and the fusion of it with the conventional adult face recognition engine also works well on the overall data set containing both baby and adult faces.