项目名称: 在轨视频图像特征提取与压缩关键技术研究
项目编号: No.91538111
项目类型: 重大研究计划
立项/批准年度: 2016
项目学科: 计算机科学学科
项目作者: 张史梁
作者单位: 北京大学
项目金额: 68万元
中文摘要: 随着卫星图像视频分辨率的持续提高,如何对视频图像数据进行实时、高效传输已成为新一代卫星技术面临的重要问题之一。针对该问题,本课题对在轨端视频图像特征提取和压缩新算法开展研究。具体研究内容包括:(1)紧凑局部特征提取、(2)高判别力视觉基元提取、(3)紧凑语义特征提取、(4)视频特征压缩编码。通过融合并压缩图像区域的局部视觉信息和空间结构信息构建紧凑、高判别力的局部特征;通过发掘感兴趣物体中的固定、可重复的视觉模式来提取视觉基元码本;利用深度学习模型提取图像高层语义信息;针对视频提取多层次的特征,并设计相应的帧间压缩算法。提取并压缩多层次的互补特征有望在实现特征数据高效传输的同时保证特征的通用性和对视频图像内容的判别能力,提升后期数据分析性能。本课题预期为解决关键科学问题:“空间信息网络对海上目标连续观测基础理论与关键技术”做出贡献,并为空间信息网络研究提供理论、技术和方法支持。
中文关键词: 特征提取;特征描述;图像特征;局部特征;全局特征
英文摘要: The ever increasing resolution of remotely-sensed image and video has made the efficient image and video data transmission through wireless network a challenging issue. To conquer this issue, this project proposes to extract and compress visual features on the remote side and transmit the visual features instead of the raw images and videos. We mainly focus on the following tasks: (1) compact local feature extraction, which is finished by fusing two complementary clues: local visual clue and spatial configuration; (2) highly discriminative visual pattern extraction by exploiting the repeatable and discriminative local visual patterns in objects of interest; (3) compact semantic feature extraction using deep learning algorithms; (4) video feature extraction and inter-frame feature compression. Extracting and compressing the above multi-level complementary features is expected to achieve more efficient and effective visual content analysis, while maintaining efficient real-time data transmission. This study is expected to contribute to the key research project: The Basic Theory and Key Technology of Spatial Information Network for Continuous Surveillance of Maritime Targets, and provide theoretical, technical and methodological support for the research of spatial information network .
英文关键词: Feature Extraction;Feature Description;Image Feature;Local Feature;Global Feature