Dr. Xinbo Gao's Research


As the name suggests, VIPs Lab focuses on the theoretical research and system development of video and image (visual information) processing (perception) based on the machine learning and deep neural networks. In addition, we also do some research on communications, especially the networking and Transmission on the high attitude platform. The detailed research topics include machine learning, heterogeneous image reconstruction or synthesis, image quality assessment, intelligent visual perception and remote sensing image processing and analysis. Some topics are given as follows.

Machine Learning New Methods: Deep Learning, Transfer Learning, Reinforcement Learning etc
New Models: Probabilistic Graph Model, Random Forest, HMM, GMM etc
New Representation: Non-negative Matrix Decomposition, Glocal (Global-local) features, etc
Heterogenerous Image Reconstruction Image Super-resolution Reconstruction from Multi-frame Images or Video Sequence
Single Image Super-resolution Reconstruction based on Dictionary Learning or GAN-like networks
High-resolution Face Reconstruction from Clips of Video Surveillance
Image Synthesis from Photo to Sketch
Image Synthesis from Sketch to Photo
Frontal Face Synthesis from Multi-view Faces
Please access the HIT Group's Webpage
Image Quality Assessment Full-reference Image or Video Quality Assessment
Reduced-reference Image or Video Quality Assessment
No-reference Image or Video Quality Assessment
Image Quality Assessment Based on fidelity, intelligibility, and aesthetics
Please access the IQA Webpage
Intelligent Image Perception Content-based Image Retrieval and Applications to Wisdom Tourism etc
Scene Perception based on New Topic Models
Pattern Recognition: Face Recognition, Character Recognition etc
Remote Sensing Image Processing & Analysis Remote Sensing Imaginary (RSI) based on Satellite, Unmanned Aerial Vehicle etc
3D Reconstruction based on Unmanned Aerial Vehicle
RSI Enhancement (Defogging, Dehazing) and Quality Assessment
Change Detection, Scene Perception and Object Recognition
Other Applications to Medical Image Analysis, Aurora Image Analysis etc
Networking and Communications on HAPS Networking on Demand for 3D Ad hoc
Cooperative Transmission: Intellectual routing, Network coding
Feedback based Network Dynamic Programming

In the past, we did some research on the computational intelligence, visual information processing, analysis and understanding, and their applications on natural images, medical images, biometrics, security and communications. Also, we have developed some application system, especially software packages, such as PACS, medical image visualization system, 3G multimedia massaging service system and food safety detection system.

  Machine Learning: Algorithms and Theoretical Frameworks

As an active field of artificial intelligence (AI), machine learning is concerned with the task to make the computer obtain the intelligence of human beings by learning or computation, for examples, dimensionality reduction, feature extraction, classification, clustering, regression or fitting. According to the degree of dependence on the labeled samples, the machine learning is divided into three categories, supervised, unsupervised and semi-supervised learning. There are many interesting topics in this direction:

  Sample Selection and Feature Selection: Bootstarp, Bagging, Boosting-like algorithms, Relevance feedback

  Dimensionality Reduction:  Linear or nonlinear methods, Kernel PCA, LAD and their variations, GP-LVM, etc

  Supervised Learning: Fuzzy NN, Kernel-based algorithms, Incremental Learning, etc

  Unsupervised Learning: Cluster tendency, Fuzzy cluster analysis and Cluster validity, etc

  Semi-supervised Learning: LNP-based algorithm, Pairwise constraint, Transfer learning, etc

Note: During the past ten years, I devoted myself to the research of cluster analysis. Cluster analysis is one of multivariate statistical analysis methods, and a branch of unsupervised pattern recognition. Its goal is to partition an unlabeled sample set into clusters such  that the homogenous samples are classified into the same cluster, and the inhomogeneous samples are classified into different clusters. As an effective analysis tool, cluster analysis has been widely used in image processing, computer vision, pattern recognition, fuzzy control and data mining.

  Computational Intelligence: Algorithms and Theoretical Frameworks

Computational intelligence is an offshoot of artificial intelligence. As an alternative to GOFAI (good old-fashioned artificial intelligence) it rather relies on heuristic algorithms such as in fuzzy systems, neural networks and evolutionary computation. In addition, computational intelligence also embraces techniques that use Swarm intelligence, Fractals and Chaos Theory, Artificial immune systems, Wavelets, etc. Computational intelligence combines elements of learning, adaptation, evolution and Fuzzy logic (rough sets) to create programs that are, in some sense, intelligent. Computational intelligence research does not reject statistical methods, but often gives a complementary view (as is the case with fuzzy systems). Artificial neural networks is a branch of computational intelligence that is closely related to machine learning.

  Fuzzy System:  Fuzzy Logic, Rough Sets, Vague Sets, Type-II Fuzzy, Granular computing, etc

  Neural Networks: BP-like, RBF, LVQ, ART, SVM, Ying-Yang machine, etc

  Natural-Inspired Algorithms: Genetic algorithm, Evolutionary programming, Artificial immune iystem, Clonal selection algorithm

  Swarm Intelligence: Particle swarm optimization, Ant colony Algorithm, and Particle filtering

  Sparsity Representation and Compressed Sensing: Multiscale geometric analysis, Dictionary learning, etc

    Visual Information Processing, Analysis and Understanding

With the rapidly development of multimedia technology and internet, image and video becomes a popular and common media to transfer information. Hereby, image and video encoding (for storage or transport), processing (enhancement, restoration, and segmentation) and analysis (feature extraction based on color, region, edge, and texture)  attract more and more attention of researches. Content-based information retrieval, digital watermarking, and image and video quality assessment  become the most important research issues. Our research interests include:

  Encoding: JPEG2000 for image, H.264, MPEG-4, MPEG-7 for video, H.323/SIP video conference system

  Video Enhancement: Super-resolution reconstruction, De-interlacing, Frame rate up-conversion, Deblocking, In-painting, etc

  Digital Watermarking: Robust and invertible watermarking, Image and video fingerprint, steganography, etc

  Quality assessment for visual information: Perception-based  assessment methodology, Metrics and Databases

  Segmentation: Spatial segmentation, Temporal segmentation for video, Spatio-temporal segmentation, /Level Sets

  Content-based Information Retrieval:  Search engine, Content analysis, Automatic annotation, re-ranking, etc

  Situation Analysis: Semantic analysis for News or Soccer program,  Aurora classification, Situation association

  Video surveillance: Face detection, modeling (AAM), tracking, face sketch-based retrieval and recognition

  Heterogeneous Image Synthesis and Recognition:  photo-sketch-cartoon, Near infrared-visible light-thermal infrared

  Online Detection System: X-Ray detection for food security, Roller detection for paper-making, illegal information detection for MMS

    Medical Image Processing and Analysis System

As an important application field of image engineering, medical image processing attracts much more of our research interest in recent years. With the theory and practice of image processing, we hope to speed up the progress of digital hospital and computer-aided diagnosis. This research direction mainly focuses on the development of medical image processing system (Software).

  Protocol Parsing: DICOM3.0, HL-7 and related protocol

  Scientific Visualization: 3D Reconstruction (surface and volume rendering), display, measurement and analysis

  PACS: Picture archive and communication systems, Viewer, Interactive analysis

  CAD: Computer-aided detection and diagnosis (Mammography, Lung)

  IGS: Image-based guidance for surgery, Multi-modality fusion

    Near Earth Space Information Grid

Near earth space information grid is a 3D Ad Hoc networks with multiple layers, which has the following features: (i) it can work for a relatively long time with prompt response; (ii) it is able to perform the surveillance and detection continuously; (iii) it can generate and change the network topology on demand. Hence, it would be widely applied to the circumstances like military detection, communication relay, anti-terrorism and disaster rescue. Based on the above characteristics, this project examines the 3D Ad hoc network design and cooperative transmissions. Through network optimization, and topology adjustment, we achieve the optimal network under service demand to transmit the information efficiently with high reliability at the expense of minimum cost. Specifically, we investigate the following three key issues:

  3D Ad Hoc Network Design on Demand: Network modeling, Multi-objective optimization for networks

  Cooperative Transmission:  Intellectual routing, Network coding, Physical layer coding and cooperation

  Feedback based Network Dynamic Programming: Performance feedback, Topological adjustment

We explore the fundamental scientific issues from the practical applications, which are related to the network information theory. This study becomes challenging and insightful, when we consider it from a new perspective, new applications, and new circumstances. This project will produce the new theory, novel techniques, which lay out the foundations for the near earth space information grid networks.


  [ Home | Resume | Grants | Publications | Schedule | Research | Teaching | VIPSL | SEE | Xidian]

Last Modified: 2023-08-20