WHU-CVRS

News

On July 4th, 2015, postgraduate academic forum on computer vision was held successfully. The forum was hosted by WHU-CVRS Lab and Prof. Liangpei Zhang’s computer vision team from State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing.

First, Prof. Jian Yao from School of Remote Sensing and Information Engineering, who was also the supervisor in WHU-CVRS Lab presided over the opening ceremony and welcomed the guests. Professor Xiangyun Hu, Dr. Yi Zhang from School of Remote Sensing and Information Engineering, Professor Chunxia Xiao, Jing Huang from Computer School, Professor Wen Yang from Electronic Information School and Professor Guisong Xia from State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing attended the meeting.

After the opening ceremony, the computer vision team from State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing led by Prof. Liangpei Zhang and WHU-CVRS Lab led by Prof. Jian Yao reported on their current research and result.

Fan Hu, a Master-PhD student from the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing gave his first academic report, which was entitled “Deep learning in high-resolution remote sensing image understanding”. In his report, he briefly introduced the basic concept of deep learning and how the convolutional neural network can be applied in object recognition and high-resolution remote sensing image understanding. Teachers and students present were interested in deep learning and asked a lot of questions. Fan Hu answered their questions patiently.

Tong He, a postgraduate from School of Remote Sensing and Information Engineering, was the second reporter. The title of his report was “Scene text detection and recognition” where he introduced some applications of deep learning in text detection.

The next student was Li Li, a postgraduate from School of Remote Sensing and Information Engineering. The title of his report was “Automatic panoramic image mosaic system” where he introduced the automatic panoramic image mosaic system invented by CVRS Lab. The system mainly included image distortion correction, searching for the best fight wiring and unifying light and color, etc. The functions above can also be extended to aerial image.

Since 3D image reconstruction is an important problem in computer vision and photogrammetry, Nan Xue, a MD-PHD student delivered his report entitled “Variational methods for 3D reconstruction”. In his report, first, he introduced the procedure of 3D reconstruction and the recovery of surface model. Based on variational calculus, he interpreted the variational model and its solution for surface reconstruction.

Mi Zhang, a Master-PhD postgraduate from School of Remote Sensing and Information Engineering integrated his experience in CVPR'2015 Conference and his paper “Line-Based Multi-Label Energy Optimization for Fisheye Image Rectification and Calibration” in CVPR'2015. He introduced the application of Multi Label Energy Optimization(MLEO) in Fisheye image rectification and calibration. He also summarized the development in 3D reconstruction prompted in CVPR'2015 Conference.

The next reporter was Xiaohu Lu, a postgraduate from School of Remote Sensing and Information Engineering, he mainly introduced some algorithms of edge detection and line segment extraction.

Feng Yang, a postgraduate from the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing delivered his report entitled “Query the potential of spatial texture features in dynamic texture recognition”. In his report, the dynamic texture is regarded as a bunch of static images and the recognition can be done simply by clustering spatial texture features of single frame image.

Kai Li, a postgraduate from School of Remote Sensing and Information Engineering introduced line segment matching and how the line segment matching can be used in 3D reconstruction.

The workload of visual interpretation of high resolution remote sensing data heavy and the threshold is high, which makes it a huge obstacle in remote sensing data processing. To solve the problem, Zifeng Wang, a postgraduate from the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing introduced a semi-automatic algorithm for high resolution remote sensing image based on active clustering. The basis of the algorithm is spectral clustering; the high efficient pair-supervision can be obtained by self-learning and human-computer interaction; then gradually revise the graph structure for spectrum clustering and optimize the clustering results. The algorithm ensures higher accuracy while reducing manual work.

At the end of this forum, teachers and students who delivered reports took a photo to mark the occasion.