视频图像分割中自适应超像素实时生成问题研究

Superpixel segmentation, as a fundamental commonality problem in the fields of computer vision and image, video processing, is a key issue and difficulty that needs to be solved urgently. This project with an application background of the real-time requirements of surveillance video intelligence analysis in social public safety, thoroughly studies the problems in the real-time superpixel generation of video image adaptively. The project focuses on the theories and algorithms about the intrinsic property description and superpixel feature extraction, generating image superpixel with adaptive multi-scale, and generation of adaptive real-time superpixels for video frames. The main objectives of this project are: to propose a new multi-level edge extraction operator based on the mathematical mechanism of fuzzy clustering, to establish a model for determining superpixel adaptively based on the analysis of the complexity of images, to propose a real-time superpixel generation algorithm for video frame images based on time domain information and non-local similarity, and to achieve the basic requirements of accuracy and real-time performance in the superpixel generation method. The research contents of this project include theories, techniques, and algorithms for adaptive superpixel generation. The completion of the project will provide new theories and a series of effective methods for solving the problem of real-time generation of adaptive superpixels for video images. It will also provide new approaches for research and application in the above areas. Therefore, the project has a very good theoretical significance and application value.

超像素分割是计算机视觉、视频图像处理和虚拟现实等领域的共性基础问题，也是亟需解决的关键问题和难点问题。本项目以社会公共安全中监控视频智能分析应用的实时性需求为背景，深入研究视频图像自适应超像素实时生成中需解决的问题。研究重点是：超像素的内在性质描述及特征提取，图像的自适应多尺度超像素生成和视频图像超像素实时生成的理论和算法。本项目的主要目标是：基于模糊聚类的数学机理提出新的多层次边缘提取算子；在分析度量图像复杂程度的基础上建立超像素尺度自适应确定模型；基于时域信息与非局部相似性提出视频帧图像的超像素实时生成算法，进而达到超像素生成方法中对准确性与实时性的基本要求。本项目的研究内容包括自适应超像素生成方面的理论、技术和算法。项目的完成将为解决视频图像自适应超像素的实时生成问题提供新的理论和一系列有效方法，也为上述等领域研究及应用提供新的途径和方法。因此，项目具有很好的理论意义和应用价值。

超像素分割是计算机视觉、视频图像处理和虚拟现实等领域的共性基础问题，也是亟需解决的关键问题和难点问题。本项目以视频图像智能分析应用的实时性需求为背景，深入研究视频图像自适应超像素实时生成中需解决的问题。研究内容主要包括超像素的内在性质描述与特征提取、图像的自适应多尺度超像素生成和视频图像超像素实时生成的理论和算法。项目围绕既定研究内容，尤其在图像去噪、平滑、超像素算法的实时性、鲁棒性与自适应性、以及图像块的快速查找方面提出了一系列的理论方法和实用技术。相关成果以论文形式在国内外重要学术期刊上共发表10篇（其中SCI检索论文9篇，EI检索论文1篇），获得授权发明专利2项。所取得的研究成果为视频图像分割等领域提供了新的理论和方法。